Method and terminal for measuring DCR of battery of electric vehicle

Abstract

A method and terminal for measuring the DCR of a battery of an electric vehicle. The method comprises: S1, on the basis of a standard directive-current-resistance calculation formula, performing multiple instances of calculation to obtain first directive current resistances within different time periods; S2, on the basis of an MAPE indicator, measuring fluctuations between average values and the first directive current resistances obtained by means of the multiple instances of calculation; S3, removing measurement errors from historical first directive current resistances, and using an LSTM model to perform regression fitting, so as to obtain second directive current resistances after the errors are removed; and S4, on the basis of the first directive current resistances and the second directive current resistances, using a convolutional network to train a directive-current-resistance estimation model, and subsequently calculating a real-time directive current resistance value of a vehicle battery in real time by means of the directive-current-resistance estimation model. By means of the method and the terminal, the precondition that a vehicle power source needs to be turned off during original calculation of the directive current resistance of the vehicle battery is removed, and an environmental error and a random error in directive current resistances are also removed, such that the directive current resistance value calculated by using a model method has a smaller fluctuation, and the result is more reliable and stable.

Description

A DCR detection method and terminal for electric vehicle batteries Technical Field

[0001] The present invention relates to the technical field of electric vehicle batteries, and in particular to a DCR detection method and terminal for an electric vehicle battery. Background Art

[0002] Due to the continuous reduction of traditional energy and its pollution to the environment, the utilization and development of new energy have been raised to a new level. The popularity of electric vehicles is increasing. Batteries are the core components of electric vehicles, and the direct current resistance (DCR) of batteries is particularly important. Improving the stability of DC resistance detection is of great significance to the safety assessment of vehicle batteries.

[0003] Existing testing schemes usually fail to fully meet standard operating conditions during daily vehicle charging. Multiple test results of the DC internal resistance of the same vehicle will produce large fluctuations, which will greatly interfere with the assessment of the vehicle battery safety status and cause inaccurate test results. Technical issues

[0004] The technical problem to be solved by the present invention is to provide a DCR detection method and terminal for an electric vehicle battery, thereby reducing environmental errors and random errors and making the DCR calculation results more reliable and stable. Technical Solutions

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for detecting DCR of an electric vehicle battery comprises the following steps:

[0007] S1. Calculate the first DC internal resistance for different time periods multiple times based on a standard DC internal resistance calculation formula;

[0008] S2. Measuring the fluctuation of the first DC internal resistance obtained by multiple calculations and the average value based on the MAPE index;

[0009] S3, removing the historical detection error of the first DC internal resistance, and performing regression fitting using the LSTM model to obtain the second DC internal resistance after the error is removed;

[0010] S4. Using a convolutional network to train a DC internal resistance estimation model based on the first DC internal resistance and the second DC internal resistance, and subsequently calculating a real-time DC internal resistance value of the vehicle battery in real time using the DC internal resistance estimation model.

[0011] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0012] An electric vehicle battery DCR detection terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When executing the computer program, the processor implements the following steps:

[0013] S1. Calculate the first DC internal resistance for different time periods multiple times based on a standard DC internal resistance calculation formula;

[0014] S2. Measuring the fluctuation of the first DC internal resistance obtained by multiple calculations and the average value based on the MAPE index;

[0015] S3, removing the historical detection error of the first DC internal resistance, and performing regression fitting using the LSTM model to obtain the second DC internal resistance after the error is removed;

[0016] S4. Using a convolutional network to train a DC internal resistance estimation model based on the first DC internal resistance and the second DC internal resistance, and subsequently calculating a real-time DC internal resistance value of the vehicle battery in real time using the DC internal resistance estimation model. Beneficial effects

[0017] The beneficial effect of the present invention is that: by first using the existing standard DC internal resistance calculation formula to calculate the DC internal resistance of the vehicle battery for multiple times in different time periods to obtain multiple groups of multiple first DC resistance values, which are used as the basis for subsequent training of the DC resistance estimation model, and before training the model, the calculated first DC resistance value is first measured based on the MAPE indicator to measure the fluctuation between the calculated value and the average value of multiple calculations, and then the error is removed and regression fitting is performed using the LSTM model to obtain the calculated value after the error is removed - the second DC internal resistance; finally, based on the convolutional network, the calculated initial value first DC resistance value and the second DC resistance after the error is removed are trained to obtain a DC internal resistance estimation model for subsequent real-time DC internal resistance value calculation of the vehicle battery, which can effectively eliminate the pre-condition in the original standard calculation method that requires turning off the vehicle power and leaving the vehicle stationary for a period of time, while eliminating environmental errors and random errors in the calculation process, making the DCR calculation result more reliable and stable. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] FIG1 is a main flow chart of a method for detecting DCR of an electric vehicle battery according to an embodiment of the present invention;

[0019] FIG2 is a schematic structural diagram of a DCR detection terminal for an electric vehicle battery according to an embodiment of the present invention.

[0020] Description of labels:

[0021] 1. A DCR detection terminal for an electric vehicle battery; 2. A memory; 3. A processor. Modes for Carrying Out the Invention

[0022] To illustrate the technical content, achieved objectives and effects of the present invention in detail, the following description is given in conjunction with the embodiments and accompanying drawings.

[0023] Before that, the English abbreviations appearing in the text are explained as follows:

[0024] MAPE indicator: Mean Absolute Percentage Error, also known as the average absolute percentage error, is a measurement method commonly used to evaluate the error between the predicted value and the actual value. It is an important indicator for measuring the accuracy of the predicted value.

[0025] LSTM model: Long Short-Term Memory, is a deep learning model used to process sequence data. It can effectively solve long-term dependency problems and is widely used in natural language processing, speech recognition, time series prediction and other fields.

[0026] Referring to FIG. 1 , a method for detecting DCR of an electric vehicle battery includes the following steps:

[0027] S1. Calculate the first DC internal resistance for different time periods multiple times based on a standard DC internal resistance calculation formula;

[0028] S2. Measuring the fluctuation of the first DC internal resistance obtained by multiple calculations and the average value based on the MAPE index;

[0029] S3, removing the historical detection error of the first DC internal resistance, and performing regression fitting using the LSTM model to obtain the second DC internal resistance after the error is removed;

[0030] S4. Using a convolutional network to train a DC internal resistance estimation model based on the first DC internal resistance and the second DC internal resistance, and subsequently calculating a real-time DC internal resistance value of the vehicle battery in real time using the DC internal resistance estimation model.

[0031] From the above description, it can be seen that the beneficial effect of the present invention is that: by first using the existing standard DC internal resistance calculation formula to calculate the DC internal resistance of the vehicle battery in different time periods multiple times to obtain multiple groups of multiple first DC resistance values, as the basis for subsequent training of the DC resistance estimation model, and before training the model, the calculated first DC resistance value is first measured based on the MAPE indicator to measure the fluctuation between the calculated value and the average value of multiple calculations, and then the error is removed and regression fitting is performed using the LSTM model to obtain the calculated value after the error is removed - the second DC internal resistance, and finally based on the convolutional network, the calculated initial value first DC resistance value and the second DC internal resistance after the error is removed are trained to obtain a DC internal resistance estimation model for subsequent vehicle battery real-time DC internal resistance value calculation, which can effectively eliminate the pre-condition in the original standard calculation method that requires turning off the vehicle power and standing still for a period of time, while eliminating environmental errors and random errors in the calculation process, making the DCR calculation result more reliable and stable.

[0032] Furthermore, the standard DC internal resistance calculation formula in step S1 is specifically:

[0033] (1);

[0034] Wherein, DCR1 is the first DC internal resistance, U1 and I1 are respectively the total voltage value and current value of the 10th second when the battery constant current output is 0.1I (A) for 10 seconds, and U2 and I2 are respectively the total voltage value and current value of the 10th second when the battery constant current output is 1 (A) for 10 seconds.

[0035] Furthermore, the step S2 is specifically as follows:

[0036] S21, collecting the first DC internal resistance DCR1 obtained by historical calculation, and retaining the DCR1 detected more than 5 times;

[0037] S22. Based on the MAPE indicator, the fluctuation of the first DC internal resistance DCR1 obtained by multiple calculations and the average value is measured according to formula (2):

[0038] (2);

[0039] Among them, Mope modify is the adjusted MAPE indicator, is the average value of the DC resistance when the i-th group is tested, m is the number of DCR1 groups, x ij is the DCR1 calculated for the jth test in the i-th group of tests, and n is the number of DCR1 tests in the i-th group.

[0040] From the above description, we can see that the DCR1 calculated when the number of detections is greater than 5 is retained to ensure the diversity of the number of samples in the training model; at the same time, the MAPE indicator can effectively measure the fluctuation between the calculated value and the average value, ensuring the determination of the loss function used in the subsequent regression fitting to avoid underfitting and overfitting.

[0041] Furthermore, the step S3 is specifically as follows:

[0042] S31. Split the sequence data of the first DC internal resistance calculated based on historical vehicle detection into a trend term, a seasonal term, and a residual term using an STL sequence decomposition algorithm, and remove the residual term.

[0043] S32. Use an LSTM model to perform regression fitting on the sequence data of the first DC internal resistance after removing the residual term to obtain a second DC internal resistance after removing the error;

[0044] The fitting loss function loss is based on the MAPE indicator, and the loss value is controlled to be the Mape calculated in step S2. modify value.

[0045] As can be seen from the above description, by first removing the residual term of the calculated first DC internal resistance and then using the LSTM model for regression fitting to obtain the second DC internal resistance, not only does this expand the diversity of samples, but the removal of the erroneous current resistance value also further ensures the generalization ability of the DC internal resistance estimation model obtained through subsequent training, ensuring the stability and reliability of the subsequent calculation of the real-time DC internal resistance value.

[0046] Furthermore, the step S4 is specifically as follows:

[0047] S41, dividing the historical first DC internal resistance and the second DC internal resistance after error removal into a training set and a test set;

[0048] S42: Using the first DC internal resistance in the training set to construct features of a convolutional network, using the second DC internal resistance in the training set as a training target of the convolutional network, and training a DC internal resistance estimation model;

[0049] S43. Testing the accuracy of the DC internal resistance estimation model using the first DC internal resistance and the second DC internal resistance in the test set, and performing regularization, data augmentation, gradient clipping, and learning rate adjustment to obtain a revised DC internal resistance estimation model;

[0050] S44 , subsequently calculating the real-time DC internal resistance value of the vehicle battery in real time using the revised DC internal resistance estimation model.

[0051] From the above description, we can see that using a convolutional network to train a model has a supervisory and correction effect during the model training process, ensuring that the DCR value calculated using the model subsequently fluctuates less and the results are more reliable and stable.

[0052] Referring to FIG2 , an electric vehicle battery DCR detection terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0053] S1. Calculate the first DC internal resistance for different time periods multiple times based on a standard DC internal resistance calculation formula;

[0054] S2. Measuring the fluctuation of the first DC internal resistance obtained by multiple calculations and the average value based on the MAPE index;

[0055] S3, removing the historical detection error of the first DC internal resistance, and performing regression fitting using the LSTM model to obtain the second DC internal resistance after the error is removed;

[0056] S4. Using a convolutional network to train a DC internal resistance estimation model based on the first DC internal resistance and the second DC internal resistance, and subsequently calculating a real-time DC internal resistance value of the vehicle battery in real time using the DC internal resistance estimation model.

[0057] As can be seen from the above description, the beneficial effect of the present invention is that: based on the same technical concept, in conjunction with the above-mentioned electric vehicle battery DCR detection method, an electric vehicle battery DCR detection terminal is provided, by first using the existing standard DC internal resistance calculation formula to calculate the vehicle battery DC internal resistance at different time periods multiple times to obtain multiple groups of multiple first DC resistance values, which serve as the basis for subsequent training of the DC resistance estimation model. Before training the model, the calculated first DC resistance value is first measured based on the MAPE indicator to measure the fluctuation between the calculated value and the average value of multiple calculations, and then the error is removed and regression fitting is performed using the LSTM model to obtain the calculated value after error removal - the second DC internal resistance. Finally, based on the convolutional network, the calculated initial value first DC resistance value and the second DC resistance after error removal are trained to obtain a DC internal resistance estimation model for subsequent vehicle battery real-time DC internal resistance value calculation. This can effectively eliminate the pre-condition in the original standard calculation method that requires the vehicle power to be turned off and the vehicle to be stationary for a period of time. At the same time, the environmental error and random error in the calculation process are eliminated, making the DCR calculation result more reliable and stable.

[0058] Furthermore, the standard DC internal resistance calculation formula in step S1 is specifically:

[0059] (1);

[0060] Wherein, DCR1 is the first DC internal resistance, U1 and I1 are respectively the total voltage value and current value of the 10th second when the battery constant current output is 0.1I (A) for 10 seconds, and U2 and I2 are respectively the total voltage value and current value of the 10th second when the battery constant current output is 1 (A) for 10 seconds.

[0061] Furthermore, the step S2 is specifically as follows:

[0062] S21, collecting the first DC internal resistance DCR1 obtained by historical calculation, and retaining the DCR1 detected more than 5 times;

[0063] S22. Based on the MAPE indicator, the fluctuation of the first DC internal resistance DCR1 obtained by multiple calculations and the average value is measured according to formula (2):

[0064] (2);

[0065] Among them, Mope modify is the adjusted MAPE indicator, is the average value of the DC resistance when the i-th group is tested, m is the number of DCR1 groups, x ij is the DCR1 calculated for the jth test in the i-th group of tests, and n is the number of DCR1 tests in the i-th group.

[0066] From the above description, we can see that the DCR1 calculated when the number of detections is greater than 5 is retained to ensure the diversity of the number of samples in the training model; at the same time, the MAPE indicator can effectively measure the fluctuation between the calculated value and the average value, ensuring the determination of the loss function used in the subsequent regression fitting to avoid underfitting and overfitting.

[0067] Furthermore, the step S3 is specifically as follows:

[0068] S31. Split the sequence data of the first DC internal resistance calculated based on historical vehicle detection into a trend term, a seasonal term, and a residual term using an STL sequence decomposition algorithm, and remove the residual term.

[0069] S32. Use an LSTM model to perform regression fitting on the sequence data of the first DC internal resistance after removing the residual term to obtain a second DC internal resistance after removing the error;

[0070] The fitting loss function loss is based on the MAPE indicator, and the loss value is controlled to be the Mape calculated in step S2. modify value.

[0071] As can be seen from the above description, by first removing the residual term of the calculated first DC internal resistance and then using the LSTM model for regression fitting to obtain the second DC internal resistance, not only does this expand the diversity of samples, but the removal of the erroneous current resistance value also further ensures the generalization ability of the DC internal resistance estimation model obtained through subsequent training, ensuring the stability and reliability of the subsequent calculation of the real-time DC internal resistance value.

[0072] Furthermore, the step S4 is specifically as follows:

[0073] S41, dividing the historical first DC internal resistance and the second DC internal resistance after error removal into a training set and a test set;

[0074] S42: Using the first DC internal resistance in the training set to construct features of a convolutional network, using the second DC internal resistance in the training set as a training target of the convolutional network, and training a DC internal resistance estimation model;

[0075] S43. Testing the accuracy of the DC internal resistance estimation model using the first DC internal resistance and the second DC internal resistance in the test set, and performing regularization, data augmentation, gradient clipping, and learning rate adjustment to obtain a revised DC internal resistance estimation model;

[0076] S44 , subsequently calculating the real-time DC internal resistance value of the vehicle battery in real time using the revised DC internal resistance estimation model.

[0077] From the above description, we can see that using a convolutional network to train a model has a supervisory and correction effect during the model training process, ensuring that the DCR value calculated using the model subsequently fluctuates less and the results are more reliable and stable.

[0078] The present inventors provide a method and terminal for detecting the DCR of an electric vehicle battery, which is suitable for performing real-time and stable detection and calculation of the DC internal resistance of an electric vehicle battery, without requiring the battery to be powered off and placed in a static state. This method is described in detail below with reference to an embodiment.

[0079] Please refer to FIG1 , the first embodiment of the present invention is:

[0080] A method for detecting the DCR of an electric vehicle battery. Before using the standard DC internal resistance formula to calculate the DC internal resistance of an electric vehicle, it is necessary to adjust the power system SOC to 50% by charging and discharging the entire vehicle (including onboard electrical equipment). The vehicle is then turned off and left to stand for a period of time, such as 30 minutes, before calculating the DC internal resistance.

[0081] However, this application does not need to turn off the car power supply, but directly uses the standard DC internal resistance calculation formula to calculate the first DC internal resistance DCR1 of the car. That is, the steps of this application are shown in Figure 1:

[0082] S1. Calculate the first DC internal resistance DCR1 for multiple times in different time periods based on a standard DC internal resistance calculation formula.

[0083] S2. Measure the fluctuation of the first DC internal resistance DCR1 obtained by multiple calculations and the average value based on the MAPE index to reduce the fluctuation of the DCR result value.

[0084] S3. Remove the detection error of the historical first DC internal resistance DCR1, and use the LSTM model for regression fitting to obtain the second DC internal resistance DCR2 after the error is removed. That is, the environmental error and random error of the DC internal resistance are effectively removed.

[0085] S4. Using a convolutional network to train a DC internal resistance estimation model based on the first DC internal resistance DCR1 and the second DC internal resistance DCR2, and subsequently calculating a real-time DCR value of the vehicle battery in real time using the DC internal resistance estimation model.

[0086] That is, in this embodiment, the DC internal resistance of the vehicle battery at different time periods is calculated multiple times using the existing standard DC internal resistance calculation formula to obtain multiple groups of multiple first DC resistance values, which serve as the basis for subsequent training of the DC resistance estimation model. Before training the model, the calculated first DC resistance value is first measured based on the MAPE indicator to measure the fluctuation between the calculated value and the average value of multiple calculations, and then the error is removed and regression fitting is performed using the LSTM model to obtain the calculated value after the error is removed - the second DC internal resistance. Finally, based on the convolutional network, the calculated initial value first DC resistance value and the second DC resistance after the error is removed are trained to obtain a DC internal resistance estimation model for subsequent real-time DC internal resistance value calculation of the vehicle battery. This can effectively eliminate the pre-condition in the original standard calculation method that requires the vehicle power to be turned off and the vehicle to be stationary for a period of time. At the same time, environmental errors and random errors in the calculation process are eliminated, making the DCR calculation result more reliable and stable.

[0087] The second embodiment of the present invention is:

[0088] A method for detecting DCR of an electric vehicle battery is provided. Based on the above-mentioned embodiment 1, in this embodiment, the standard DC internal resistance calculation formula in step S1 is specifically:

[0089] (1).

[0090] Among them, DCR1 is the first DC internal resistance, U1 and I1 are the total voltage value and current value of the 10th second when the battery constant current output is 0.1I (A) for 10 seconds, respectively, and U2 and I2 are the total voltage value and current value of the 10th second when the battery constant current output is 1 (A) for 10 seconds.

[0091] The above formula (1) is used to calculate multiple groups of DCR1 values ​​as historical detection data for subsequent training model samples. In this embodiment, step S2 is specifically as follows:

[0092] S21. Collect the DC internal resistance DCR1 obtained by historical calculation and retain the DCR1 with more than 5 detection times to ensure the diversity of the number of samples for training the model.

[0093] S22. Based on the MAPE indicator, the fluctuation of the DC internal resistance DCR1 obtained by multiple calculations and the average value is measured according to formula (2):

[0094] (2).

[0095] Among them, Mope modify is the adjusted MAPE indicator, is the average value of the DC resistance when the i-th group is tested, m is the number of DCR1 groups, x ij is the DCR1 calculated for the jth test in the i-th group of tests, and n is the number of DCR1 tests in the i-th group.

[0096] That is, the use of MAPE indicators can effectively measure the fluctuation between the calculated value and the average value, ensuring the determination of the loss function used in the subsequent regression fitting to avoid underfitting and overfitting.

[0097] Meanwhile, in this embodiment, step S3 is specifically as follows:

[0098] S31. Based on the DCR1 sequence data obtained by historical vehicle inspection calculation, the sequence data is split into trend terms, seasonal terms, and residual terms using the STL sequence decomposition algorithm, and the residual is removed.

[0099] S32, use the LSTM model to perform regression fitting on the DCR1 sequence data after removing the residual term, and obtain the second DC internal resistance DCR2 after removing the error; wherein the fitting loss function loss is based on the MAPE indicator, and the loss value is controlled to the Mape calculated in step S2. modify value.

[0100] That is, by first removing the residual term of the calculated DCR1 and then using the LSTM model for regression fitting to obtain the DCR2 value, not only does this expand the diversity of samples, but the removal of the erroneous current resistance value also further ensures the generalization ability of the DCR estimation model obtained through subsequent training, ensuring the stability and reliability of subsequent DCR calculations.

[0101] Among them, it is worth mentioning that in deep learning, when performing gradient calculations, an indicator is needed to measure the degree of difference between the calculated value and the actual target value. The indicator for measuring this difference is called the loss function. In this embodiment, the MAPE indicator is used as a method for calculating loss, which can effectively measure the fluctuation between the calculated value and the average value, and ensure the determination of the loss function used for subsequent regression fitting. However, the loss function loss value will continue to increase with the increase in the number of training iterations. When the loss value is too large, it will cause regression underfitting, making the subsequent training model inaccurate. If the loss value is too small, it will cause regression overfitting, making the subsequent training model generalization ability weak. Therefore, in this embodiment, the loss value is trained based on the MAPE indicator to the Mope obtained by removing the fluctuation in the above formula (2). modify Nearby, so that the loss value has a suitable value, neither too large nor too small, to avoid underfitting and overfitting.

[0102] In addition, in this embodiment, step S4 is specifically as follows:

[0103] S41, dividing the historical first DC internal resistance DCR1 and the second DC internal resistance DCR2 after error removal into a training set and a test set;

[0104] S42. Use DCR1 in the training set to construct the features of the convolutional network, use DCR2 in the training set as the training target of the convolutional network, and train the DC internal resistance estimation model;

[0105] S43. Use DCR1 and DCR2 in the test set to test the accuracy of the DCR estimation model, and add regularization, data enhancement, gradient clipping, and learning rate adjustment to obtain a revised DC internal resistance estimation model;

[0106] S44. Subsequently, the real-time DCR value of the vehicle battery is calculated in real time using the corrected DC internal resistance estimation model.

[0107] That is, using a convolutional network to train the model has a supervisory and correction role during the model training process, ensuring that the DCR value calculated using the model later fluctuates less and the results are more reliable and stable.

[0108] Please refer to FIG2 , the third embodiment of the present invention is:

[0109] An electric vehicle battery DCR detection terminal 1 includes a memory 2, a processor 3, and a computer program stored in the memory 2 and executable on the processor 3. In this embodiment, when the processor 3 executes the computer program, the steps in the above-mentioned embodiment 1 or embodiment 2 are implemented.

[0110] In summary, the present invention provides a method for detecting the DCR of an electric vehicle battery. Since the accuracy of the existing DC internal resistance detection scheme depends on the pre-conditions, if the pre-conditions are not fully met, the DC internal resistance calculated by detection will produce a large error. Therefore, the improved scheme provided by the present invention eliminates the original pre-conditions and the environmental error and random error of the DC internal resistance. The DCR value calculated using the model method has smaller fluctuations and the result is more reliable and stable.

[0111] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's description and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for detecting DCR of an electric vehicle battery, characterized in that: Includes steps: S1. Calculate the first DC internal resistance at different time periods multiple times based on a standard DC internal resistance calculation formula; S2, measuring the fluctuation of the first DC internal resistance obtained by multiple calculations and the average value based on the MAPE index; S3, removing the historical detection error of the first DC internal resistance, and using the LSTM model for regression fitting to obtain the second DC internal resistance after the error is removed; S4. Using a convolutional network to train a DC internal resistance estimation model based on the first DC internal resistance and the second DC internal resistance, and subsequently calculating a real-time DC internal resistance value of the vehicle battery in real time using the DC internal resistance estimation model.

2. A method for detecting DCR of an electric vehicle battery according to claim 1, characterized in that: The standard DC internal resistance calculation formula in step S1 is specifically: （1）； Among them, DCR1 is the first DC internal resistance, U1 and I1 are respectively the total voltage value and current value of the 10th second when the battery constant current output is 0.1I (A) for 10s, and U2 and I2 are respectively the total voltage value and current value of the 10th second when the battery constant current output is I (A) for 10s.

3. A method for detecting DCR of an electric vehicle battery according to claim 2, characterized in that: The step S2 is specifically as follows: S21, collecting the first DC internal resistance DCR1 obtained by historical calculation, and retaining the DCR1 detected more than 5 times; S22. Based on the MAPE indicator, the fluctuation of the first DC internal resistance DCR1 obtained by multiple calculations and the average value is measured according to formula (2): （2）； Among them, Mope modify is the adjusted MAPE indicator, is the average value of the DC resistance of the i-th group, m is the number of groups of DCR1, x ij is the DCR1 calculated for the jth detection in the i-th group of detections, and n is the number of detections of DCR1 in the i-th group.

4. A method for detecting DCR of an electric vehicle battery according to claim 3, characterized in that: The step S3 is specifically as follows: S31, based on the sequence data of the first DC internal resistance calculated based on the vehicle history detection, split the sequence data into trend terms, seasonal terms and residual terms through the STL sequence decomposition algorithm, and remove the residual terms; S32, using an LSTM model to perform regression fitting on the sequence data of the first DC internal resistance after the residual term is removed, to obtain a second DC internal resistance after the error is removed; The fitting loss function loss is based on the MAPE indicator, and the loss value is controlled to be the Mape calculated in step S2. modify value.

5. A method for detecting DCR of an electric vehicle battery according to claim 4, characterized in that: The step S4 is specifically as follows: S41, dividing the historical first DC internal resistance and the second DC internal resistance after error removal into a training set and a test set; S42, using the first DC internal resistance in the training set to construct features of a convolutional network, using the second DC internal resistance in the training set as a training target of the convolutional network, and training a DC internal resistance estimation model; S43, using the first DC internal resistance and the second DC internal resistance in the test set to test the accuracy of the DC internal resistance estimation model, and adding regularization, data enhancement, gradient clipping and learning rate adjustment to obtain a revised DC internal resistance estimation model; S44. Subsequently, the real-time DC internal resistance value of the vehicle battery is calculated in real time using the corrected DC internal resistance estimation model.

6. An electric vehicle battery DCR detection terminal, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are implemented: S1. Calculate the first DC internal resistance at different time periods multiple times based on a standard DC internal resistance calculation formula; S2, measuring the fluctuation of the first DC internal resistance obtained by multiple calculations and the average value based on the MAPE index; S3, removing the historical detection error of the first DC internal resistance, and using the LSTM model for regression fitting to obtain the second DC internal resistance after the error is removed; S4. Using a convolutional network to train a DC internal resistance estimation model based on the first DC internal resistance and the second DC internal resistance, and subsequently calculating a real-time DC internal resistance value of the vehicle battery in real time using the DC internal resistance estimation model.

7. An electric vehicle battery DCR detection terminal according to claim 6, characterized in that: The standard DC internal resistance calculation formula in step S1 is specifically: （1）； Among them, DCR1 is the first DC internal resistance, U1 and I1 are respectively the total voltage value and current value of the 10th second when the battery constant current output is 0.1I (A) for 10s, and U2 and I2 are respectively the total voltage value and current value of the 10th second when the battery constant current output is I (A) for 10s.

8. The electric vehicle battery DCR detection terminal according to claim 7, characterized in that: The step S2 is specifically as follows: S21, collecting the first DC internal resistance DCR1 obtained by historical calculation, and retaining the DCR1 detected more than 5 times; S22. Based on the MAPE indicator, the fluctuation of the first DC internal resistance DCR1 obtained by multiple calculations and the average value is measured according to formula (2): （2）； Among them, Mope modify is the adjusted MAPE indicator, is the average value of the DC resistance of the i-th group, m is the number of groups of DCR1, x ij is the DCR1 calculated for the jth detection in the i-th group of detections, and n is the number of detections of DCR1 in the i-th group.

9. The electric vehicle battery DCR detection terminal according to claim 8, characterized in that: The step S3 is specifically as follows: S31, based on the sequence data of the first DC internal resistance calculated based on the vehicle history detection, split the sequence data into trend terms, seasonal terms and residual terms through the STL sequence decomposition algorithm, and remove the residual terms; S32, using an LSTM model to perform regression fitting on the sequence data of the first DC internal resistance after the residual term is removed, to obtain a second DC internal resistance after the error is removed; The fitting loss function loss is based on the MAPE indicator, and the loss value is controlled to be the Mape calculated in step S2. modify value.

10. The electric vehicle battery DCR detection terminal according to claim 9, characterized in that: The step S4 is specifically as follows: S41, dividing the historical first DC internal resistance and the second DC internal resistance after error removal into a training set and a test set; S42, using the first DC internal resistance in the training set to construct features of a convolutional network, using the second DC internal resistance in the training set as a training target of the convolutional network, and training a DC internal resistance estimation model; S43, using the first DC internal resistance and the second DC internal resistance in the test set to test the accuracy of the DC internal resistance estimation model, and adding regularization, data enhancement, gradient clipping and learning rate adjustment to obtain a revised DC internal resistance estimation model; S44. Subsequently, the real-time DC internal resistance value of the vehicle battery is calculated in real time using the corrected DC internal resistance estimation model.

Images