Method for rapidly detecting health state of retired power lithium battery

Abstract

The invention discloses a method for rapidly detecting the health state of an out-of-service power lithium battery, which belongs to the technical field of lithium battery performance detection, and comprises the following steps: obtaining a batch of out-of-service battery monomers to be detected with the same model and specification; testing the to-be-detected battery by using conventional capacity detection equipment to obtain data; processing test data of the conventional detection battery; training a classification model and an SOH estimation model by using the generated training verification data set; processing test data of the rapid detection battery; judging whether the SOH of the rapidly detected battery meets the echelon utilization requirement or not by using a classification model; the SOH estimation model is used for obtaining a fast detection battery SOH estimation value meeting the requirement; the invention provides a method for rapidly detecting the health state of a decommissioned power lithium battery, which does not need additional detection equipment and realizes efficient detection of the health state of the decommissioned power lithium battery by using conventional capacity detection equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of power battery health state detection, and particularly relates to a method for rapidly detecting the health state of retired power lithium batteries. Background Art

[0002] In recent years, with the rapid growth of the production and sales volume of electric vehicles, the installed capacity of power batteries, as their core components, has also been continuously increasing. With the increase in the service life, the capacity and performance of power batteries will gradually decay, and finally they cannot meet the usage requirements of electric vehicles and need to be retired. The traditional way to retire power batteries is to disassemble and recycle the high-value metals in them. However, most retired batteries still have great potential for cascade utilization. After screening, batteries with performance meeting the requirements can be applied to scenarios with lower battery requirements such as tourist sightseeing vehicles, three-wheeled electric bicycles, or home energy storage.

[0003] The state of health (SOH) of a battery is usually defined as the ratio of the current discharge capacity of the battery to its rated capacity, and its definition formula is:

[0004]

[0005] where C i is the current discharge capacity of the battery, and C n is the rated capacity of the battery.

[0006] During the battery screening and cascade utilization process, SOH is a key evaluation index for sorting and matching. Currently, in actual production, a complete charge-discharge test is usually used as a conventional detection method, but this method is time-consuming and energy-consuming, and has low efficiency. To improve the detection efficiency, some solutions estimate the battery SOH by combining machine learning models. In the prior art, most machine learning-based SOH detection methods (such as CN119147982A, CN119270084A, and CN119355560A, etc.) mainly focus on the online estimation of battery SOH and are difficult to be directly applied to the detection scenarios of retired battery cascade utilization. In addition, some technologies (such as CN119259515A, CN118305097A) can only achieve battery sorting and cannot provide an estimated value of SOH; while some other technologies (such as CN118655474A, CN118287405A, CN118305097A) introduce electrochemical impedance spectroscopy tests and require additional detection equipment.

[0007] When using conventional capacity detection equipment for charge and discharge testing, there are problems of high energy consumption and low efficiency, and introducing additional testing equipment will increase costs. Therefore, there is an urgent need to develop a rapid detection method for the health status of retired power lithium batteries based on existing conventional capacity detection equipment to provide an efficient and low-cost solution for the health status detection of retired batteries. Summary of the Invention

[0008] In view of the problems existing in the prior art, the present invention provides a rapid detection method for the health status of retired power lithium batteries, which does not require additional detection equipment and uses conventional capacity detection equipment to achieve efficient detection of the health status of retired batteries.

[0009] The technical solution of the present invention is realized as follows:

[0010] A rapid detection method for the health status of retired power lithium batteries includes the following steps:

[0011] S100: Obtain a batch of retired battery cells to be detected with the same model and specification, and eliminate battery cells that are obviously not suitable for cascade utilization through appearance screening, such as battery cells with defects such as deformation, bulging, or leakage.

[0012] S200: Use conventional capacity detection equipment to test the batteries to be detected to obtain data: Select 5-50% of the batteries in this batch of batteries to be detected as conventional detection batteries, use conventional capacity detection equipment to perform full capacity calibration testing and record the data; the remaining batteries are used as rapid detection batteries, and use conventional capacity detection equipment to perform rapid health status detection testing and record the data.

[0013] S300: Process the test data of the conventional detection batteries obtained in step S200: The test data of the conventional detection batteries are used for the training and verification of the classification model and the SOH estimation model.

[0014] S400: Use the training and verification data set generated after data processing in step S300 to train the classification model and the SOH estimation model to obtain the trained classification model and SOH estimation model.

[0015] S500: Process the test data of the rapid detection batteries obtained in step S200: The test data of the rapid detection batteries are used as model inputs to achieve classification and SOH estimation of the rapid detection batteries.

[0016] S600: Use the data of the rapid detection batteries after data processing in step S500 to input into the classification model to determine whether the rapid detection batteries meet the requirements for cascade utilization.

[0017] S700: For the battery data determined to meet the requirements for cascade utilization in step S600, further input it into the SOH estimation model to obtain the SOH estimated value, and use the estimated value as the rapid detection result of the health state of the battery for rapid detection.

[0018] Furthermore, in S200, when testing the battery to be detected using a conventional capacity detection device to obtain data, specifically including:

[0019] S210: Randomly select a certain proportion of the batteries to be detected in this batch as conventional detection batteries, use a conventional capacity detection device, perform constant current discharge at a rate of 0.25C - 1C until the battery voltage reaches the lower limit, let it stand for 10 minutes, perform a complete charge-discharge test and record the data;

[0020] S220: The remaining batteries in this batch of batteries to be detected are used as rapid detection batteries, use a conventional capacity detection device, perform constant current discharge at a rate of 0.25C - 1C until the battery voltage reaches the lower limit, let it stand for 10 minutes, perform a rapid detection test of the health state and record the data;

[0021] Specifically, the complete charge-discharge test includes: performing constant current charging at a rate of 0.25C - 1C until the battery voltage reaches the upper limit, then performing constant voltage charging until the cut-off current, letting it stand for 10 minutes, and performing constant current discharge at a rate of 0.25C - 1C until the battery voltage reaches the lower limit;

[0022] Specifically, the rapid detection test of the health state includes: performing constant current charging at a rate of 0.25C - 1C, and the charging time T is 30 seconds - 600 seconds;

[0023] Specifically, the recorded data includes: during the test, the changes in battery voltage, current, and capacity over time.

[0024] Furthermore, in S300, the processing of the test data of the obtained conventional detection batteries specifically includes:

[0025] S310: Generate the label of the SOH estimation model training and verification data set. In the complete charge-discharge test of the conventional detection battery, the final discharge capacity is used as the current discharge capacity C of the battery i , according to the capacity definition of the battery health state SOH, calculate the SOH of the conventional detection battery, and the formula is:

[0026]

[0027] where C i is the current discharge capacity of the battery, and C n is the rated capacity of the battery, and use the calculated SOH as the label of the SOH estimation model training and verification data set;

[0028] S320: Generate the labels for the training and validation datasets of the classification model, and set the SOH threshold SOH for determining whether a retired battery meets the requirements for cascade utilization to be 70%-90%; when the SOH of a conventional battery > Threshold SOH

[0029] SOH Threshold , its SOH meets the requirements for cascade utilization, and the classification label is recorded as 1, indicating that the requirements are met; when the SOH of a conventional battery < SOH Threshold , its SOH does not meet the requirements for cascade utilization, and the classification label is recorded as 0, indicating that the requirements are not met;

[0030] S330: Generate the input features for the training and validation datasets of the classification model and the SOH estimation model. Set the intercepted charging time T, and from the complete charge-discharge test data, intercept the constant-current charging stage data from the start of the test to the charging time T as the original data; U min is the minimum value of the voltage sequence in the intercepted data, and U max is the maximum value of the voltage sequence in the intercepted data; set the number of resampling points k, and use the battery voltage upper limit U upper and the lower limit U lower as the boundaries to generate uniformly distributed resampled voltage values; where ΔU = 0.005V;

[0031] According to the original data, for the resampling points within the range of [U min , U max , calculate the corresponding current, capacity, and time values through interpolation. For the resampling points within the ranges of [U lower , U min and [U max , U upper , the current, capacity, and time values are all set to 0, and the resampled data within the voltage range [U lower , U upper is obtained, which is a three-element sequence including current, capacity, and time; the shape of this sequence is k×3, which is used as the input features for the training and validation datasets of the classification model and the SOH estimation model; the input features correspond one-to-one with the labels of the classification model and the labels of the SOH estimation model, that is, each input feature sample corresponds to a classification label and an SOH estimation label; divide the above input features and their corresponding classification labels and SOH estimation labels into a training set and a validation set, and generate the training and validation datasets for the classification model and the SOH estimation model respectively, where the training set:validation set = 4:1.

[0032] Further, in S400, the classification model and the SOH estimation model have similar structures, both being Inception-ResNet hybrid models based on causal dilated convolutions. The specific structure includes: an input layer, a masking layer, multiple layers of causal dilated convolution Inception-ResNet modules, and an output layer; the classification model and the SOH estimation model are trained using the training and validation datasets generated in S300 to obtain the trained classification model and SOH estimation model.

[0033] Specifically, the Inception-ResNet hybrid model combines the Inception module and the Residual Network (ResNet).

[0034] Specifically, the classification model and the SOH estimation model are basically the same in structure, except for the output layer. The output layer of the classification model uses the sigmoid activation function to obtain a probability value between 0 and 1, and then outputs 0 or 1 according to the classification threshold of 0.5. The output layer of the SOH estimation model does not have an activation function and outputs a value, which is the estimated value of SOH.

[0035] Specifically, the functions of each layer of the Inception-ResNet hybrid model based on causal dilated convolutions are as follows:

[0036] The input layer receives the data input to the model, where the input to the model is a three-element sequence of current I, capacity C, and time t [(I1, C1, t1), (I2, C2, t2), …, (I k , C k , t k )].

[0037] The masking layer filters out data with a value of 0, that is, the 0 values filled in step S330 to ensure the consistency of the input feature shape.

[0038] The multiple layers of causal dilated convolution Inception-ResNet modules extract local and global features of the time series through causal convolution and dilated convolution, while maintaining causality (i.e., only using the information of the current and previous time steps); the Inception module extracts features of different scales through multi-scale convolution and fuses these features through concatenation; the residual connection adds the input features to the output of the Inception module to enhance the information flow and alleviate the problem of gradient disappearance; through multiple layers of causal dilated convolution Inception-ResNet modules, a mapping relationship between the input sequence and the output value is established.

[0039] The output layer is the final output result of the model; the classification model performs a classification task, and the output layer is 0 or 1. 0 indicates that the SOH of the corresponding battery does not meet the requirements for cascade utilization, and 1 indicates that the SOH of the corresponding battery meets the requirements for cascade utilization; the SOH estimation model performs a regression task, and the output layer is a value, that is, the estimated value of SOH.

[0040] Specifically, during the training process, the classification model uses binary cross-entropy as the loss function, and the SOH estimation model uses mean squared error as the loss function.

[0041] Further, in S500, the processing of the obtained fast-detection battery test data is similar to that in step S330, and specifically includes:

[0042] Generating input data for the classification model and the SOH estimation model for subsequent determination of whether the SOH of the fast-detection battery to be detected meets the requirements for cascade utilization and for estimating SOH; using the data in the constant-current charging stage of the state-of-health fast-detection test data as the original data, and the data in the constant-current charging stage includes voltage, current, capacity, and time;

[0043] U min is the minimum value of the voltage sequence in the original data, and U max is the maximum value of the voltage sequence in the original data; setting the number of resampling points k, and generating uniformly distributed resampled voltage values with the upper limit U upper and the lower limit U lower of the battery voltage as the boundaries; where ΔU = 0.005V;

[0044] According to the original data, for the resampling points within the range of [U min , U max , calculating the corresponding current, capacity, and time values through interpolation, and for the resampling points within the ranges of [U lower , U min and [U max , U upper , setting the current, capacity, and time values to 0, obtaining the resampled data within the voltage range of [U lower , U upper , which is a three-element sequence including current, capacity, and time; the shape of this sequence is k×3 and is used as the input for the trained classification model and SOH estimation model.

[0045] Further, in S600, using the fast-detection battery data after data processing in step S500 to input into the classification model to determine whether the fast-detection battery meets the requirements for cascade utilization; among them, the battery data determined to meet the requirements for cascade utilization continues the subsequent SOH estimation process, and the batteries determined not to meet the requirements for cascade utilization no longer continue the subsequent process and wait for scrap recycling treatment.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. The present invention utilizes effective feature engineering and an Inception-ResNet hybrid model based on causal dilated convolution. With only short-time (within one minute) charging data, it can efficiently classify retired batteries and estimate their state of health (SOH). This is faster than other machine learning SOH estimation methods that require long-time testing to provide data and is more suitable for the offline application scenario of retired battery detection.

[0048] 2. Compared with conventional full charge-discharge tests for capacity calibration, the present invention uses short-time charging data combined with machine learning methods to estimate SOH more efficiently and energy-savingly.

[0049] 3. During the testing process of the present invention, only conventional capacity detection equipment is used for battery testing, without the need to introduce new detection equipment and without incurring costs to increase equipment.

[0050] 4. The model training and validation data involved in the present invention all come from batteries of the same batch, and the method is not limited by battery types and models. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a schematic diagram of the processing flow of the battery of the present invention;

[0052] Figure 2 is a schematic diagram of the flow of the present invention;

[0053] Figure 3 is a schematic diagram of the battery testing process in the present invention;

[0054] Figure 4 is a flowchart of the processing of the conventional battery test data in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0056] Embodiment 1

[0057] As Figures 1 to 4 , a method for rapidly detecting the health state of retired power lithium batteries includes the following steps:

[0058] S100: Obtain a batch of retired battery cells of the same model and specification, and screen out the battery cells that are obviously not suitable for cascade utilization through appearance screening, such as batteries with obvious defects such as deformation, bulging or leakage. This is common knowledge in the art.

[0059] In this embodiment, a batch of A-brand ternary 18650 batteries are obtained, and the rated capacity C of the battery n is 1.85 Ah.

[0060] S200: Use a conventional capacity detection device to test the batteries to be detected to obtain data: Select 50% of the batteries to be detected in this batch, and use a conventional capacity detection device to perform a complete capacity calibration test and record the data; for the remaining batteries, use a conventional capacity detection device to perform a quick health status detection test and record the data; specifically include:

[0061] S210: Randomly select 50% of the batteries to be detected in this batch as conventional detection batteries, and use a conventional capacity detection device to perform constant current discharge at a 1C rate until the lower limit of the battery voltage, then let it stand for 10 minutes, and perform a complete charge and discharge test and record the data.

[0062] The complete charge and discharge test includes: performing constant current charging at a 1C rate until the cut-off voltage of 4.2V, that is, the upper limit of the battery voltage U upper , and then performing constant voltage charging until the cut-off current of 0.05C, let it stand for 10 minutes, and perform constant voltage charging.

[0063] Perform constant current discharge at a 1C rate until the lower limit of the battery voltage; the recorded data includes: during the test, the changes of the battery voltage, current and capacity over time.

[0064] S220: The remaining batteries in this batch of batteries to be detected are used as quick detection batteries. Use a conventional capacity detection device to perform constant current discharge at a 1C rate until the lower limit of the battery voltage, let it stand for 10 minutes, and perform a quick health status detection test and record the data; the quick health status detection test includes: performing constant current charging at a 1C rate, and the charging time T is 30 seconds; the recorded data includes: during the test, the changes of the battery voltage, current and capacity over time.

[0065] S300: Process the test data of the conventional detection batteries obtained in step S200: The test data of the conventional detection batteries are used for the training and verification of the classification model and the SOH estimation model; specifically include:

[0066] S310: Generate the label of the SOH estimation model training and verification data set. In the complete charge and discharge test of the conventional detection battery, the final discharge capacity is used as the current discharge capacity C of the battery i, according to the capacity definition of the state of health (SOH) of the battery, calculate the SOH of the regularly tested battery. The formula is:

[0067]

[0068] where C i is the current discharge capacity of the battery, and C n is the rated capacity of the battery. Use the calculated SOH as the label for the training and validation dataset of the SOH estimation model;

[0069] S320: Generate the label for the training and validation dataset of the classification model. Set the SOH threshold SOH Threshold for determining whether a retired battery meets the requirements for cascade utilization to be 70%; when the SOH of the regularly tested battery > SOH Threshold , its SOH meets the requirements for cascade utilization, and the classification label is recorded as 1, indicating that the requirements are met; when the SOH of the regularly tested battery < SOH Threshold , its SOH does not meet the requirements for cascade utilization, and the classification label is recorded as 0, indicating that the requirements are not met;

[0070] S330: Generate the input features for the training and validation datasets of the classification model and the SOH estimation model. Set the intercepted charging time T (consistent with the charging time for the rapid health detection test), and intercept the constant-current charging stage data from the start of the test to time T from the complete charge-discharge test data as the original data. The constant-current charging stage data includes voltage, current, capacity, and time.

[0071] U min is the minimum value of the voltage sequence in the intercepted data, and U max is the maximum value of the voltage sequence in the intercepted data; Set the number of resampling points k. Using the upper limit U upper of the battery voltage as 4.2V and the lower limit U lower as 3V as the boundaries, generate uniformly distributed resampled voltage values; where ΔU = 0.005V; In this embodiment, the number of resampling points k is 241.

[0072] According to the original data, for the resampling points within the range of [U min , U max , calculate the corresponding current, capacity, and time values through interpolation. For the resampling points within the ranges of [U lower , U min and [U max , U upper , set the current, capacity, and time values to 0, and obtain the voltage range [U lower , U upperThe resampled data within is a three - element sequence including current, capacity, and time; the shape of this sequence is k×3, serving as the input feature for the training and validation datasets of the classification model and the SOH estimation model; the input features correspond one - to - one with the labels of the classification model and the SOH estimation model, that is, each input feature sample corresponds to a classification label and an SOH estimation label; the above - mentioned input features and their corresponding classification labels and SOH estimation labels are divided into a training set and a validation set in a ratio of 4:1 to generate the training and validation datasets for the classification model and the SOH estimation model respectively.

[0073] S400: Use the training and validation datasets generated after data processing in step S300 to train the classification model and the SOH estimation model, obtaining the trained classification model and SOH estimation model.

[0074] The classification model and the SOH estimation model have similar structures, both being Inception - ResNet hybrid models based on causal dilated convolutions. The specific structure includes: an input layer, a masking layer, multiple causal dilated convolution Inception - ResNet modules, and an output layer; the classification model and the SOH estimation model are trained using the training and validation datasets generated in S300 to obtain the trained classification model and SOH estimation model.

[0075] The Inception - ResNet hybrid model combines the Inception module and the Residual Network (ResNet).

[0076] The classification model and the SOH estimation model are basically the same in structure, with only the output layer being different. The output layer of the classification model uses the sigmoid activation function to obtain a probability value between 0 and 1, and then outputs 0 or 1 according to the classification threshold of 0.5. The output layer of the SOH estimation model has no activation function and outputs a numerical value, which is the estimated value of SOH.

[0077] The functions of each layer of the Inception - ResNet hybrid model based on causal dilated convolutions are as follows:

[0078] The input layer receives the data input to the model, where the input to the model is a three - element sequence of current I, capacity C, and time t [(I1,C1,t1),(I2,C2,t2),…,(I k ,C k ,t k )].

[0079] The masking layer filters out the data with a value of 0, that is, the 0 values filled in step S330 to ensure the consistency of the input feature shape.

[0080] The multi-layer causal dilated convolutional Inception-ResNet module extracts local and global features of time series through causal convolution and dilated convolution while maintaining causality (i.e., only using information from the current and previous time steps); the Inception module extracts features of different scales through multi-scale convolution and fuses these features through concatenation; the residual connection adds the input features to the output of the Inception module to enhance information flow and alleviate the vanishing gradient problem. Through multiple layers of causal dilated convolutional Inception-ResNet modules, a mapping relationship is established between the input sequence and the output value.

[0081] The output layer is the final output result of the model; the classification model performs a classification task, and the output layer is 0 or 1, where 0 indicates that the corresponding battery SOH does not meet the requirements for cascade utilization, and 1 indicates that the corresponding battery SOH meets the requirements for cascade utilization; the SOH estimation model performs a regression task, and the output layer is a value, i.e., the estimated value of SOH.

[0082] During the training process, the classification model uses binary cross-entropy as the loss function, and the SOH estimation model uses mean squared error as the loss function.

[0083] S500: Process the fast detection battery test data obtained in step S200: The test data of the fast detection battery is used as the model input to achieve the classification and SOH estimation of the fast detection battery; step S330 is similar and specifically includes:

[0084] Generate the input data for the classification model and the SOH estimation model for subsequent determination of whether the SOH of the fast detection battery to be detected meets the requirements for cascade utilization and estimation of SOH; use the constant current charging stage data in the state of health fast detection test data as the original data, and the constant current charging stage data includes voltage, current, capacity, and time;

[0085] U min is the minimum value of the voltage sequence in the original data, and U max is the maximum value of the voltage sequence in the original data; set the number of resampling points k, and generate uniformly distributed resampled voltage values with the battery voltage upper limit U upper and lower limit U lower as the boundaries; according to the original data, for the resampling points within the range of [U min , U max , calculate the corresponding current, capacity, and time values through interpolation, and for the resampling points within the ranges of [U lower , U min and [U max , U upper , set the current, capacity, and time values to 0, and obtain the voltage range [U lower , U upperThe resampled data within is a ternary sequence including current, capacity, and time; the shape of this sequence is k×3, which serves as the input for the trained classification model and SOH estimation model.

[0086] S600: Input the fast-detection battery data after data processing in step S500 into the classification model to determine whether the fast-detection battery meets the requirements for cascade utilization; among them, the battery data determined to meet the requirements for cascade utilization continues with the subsequent SOH estimation process, while the batteries determined not to meet the requirements for cascade utilization no longer continue with the subsequent process and await scrapping and recycling. Among them, the classification model accurately classifies all fast-detection batteries according to SOH Threshold and is 70% accurate in classification, with an accuracy rate of 99.22%.

[0087] S700: For the battery data determined to meet the requirements for cascade utilization in step S600, further input it into the SOH estimation model to obtain the SOH estimated value, and the estimated value serves as the fast-detection result of the health status of the fast-detection battery.

[0088] In this embodiment, the average absolute error of the SOH estimation of the fast-detection batteries that meet the requirements for cascade utilization is 0.91%, and the average relative error is 0.99%; for this batch of batteries, compared with using conventional capacity detection equipment for charge and discharge testing, the detection time consumption is reduced by approximately 49.82%, and the detection energy consumption is reduced by approximately 49.55%.

[0089] The present invention provides a method for quickly detecting the health status of retired power lithium batteries, which does not require additional detection equipment and uses conventional capacity detection equipment to achieve efficient detection of the health status of retired batteries.

[0090] Compared with the prior art, the present invention has the following beneficial effects:

[0091] 1. The present invention utilizes effective feature engineering and an Inception-ResNet hybrid model based on causal dilated convolution. By using only the charging data within a short time (less than one minute), it can efficiently classify retired batteries and estimate SOH. It is faster than other machine learning SOH estimation methods that require long-time testing to provide data and is more suitable for the offline application scenario of retired battery detection.

[0092] 2. Compared with conventional full charge and discharge testing for capacity calibration, the present invention uses short-time charging data combined with machine learning methods to estimate SOH more efficiently and energy-savingly.

[0093] 3. During the testing process of the present invention, only conventional capacity detection equipment is used for battery testing, without the need to introduce new detection equipment, and there is no need to spend money to increase equipment.

[0094] 4. The model training and verification data involved in the present invention are all from the same batch of batteries, and the use of the method is not restricted by the types and models of batteries.

[0095] Example 2

[0096] In this example, in step S100, a batch of lithium iron phosphate square batteries of brand B is obtained, and the rated capacity C of the battery n is 86 Ah; in step S210, 20% of the batch of batteries to be detected are selected as conventional detection batteries; in steps S210 and S220, the constant current charging and discharging rates are both 0.25C, and the constant voltage charging cut-off current in step S210 is 0.01C; the charging time T in step S220 is 60S; the SOH Threshold in step S320 is 80%; the number k of resampling points in step S330 is 271; the remaining steps are the same as those in Example 1, and the same steps will not be repeated.

[0097] In this example, the classification model accurately classifies all the fast detection batteries according to the SOH Threshold of 80%, and the accuracy rate is 100%.

[0098] In this example, the average absolute error of the SOH estimation of the fast detection batteries meeting the requirements of cascade utilization is 2.03%, and the average relative error is 2.09%; for this batch of batteries compared with using a conventional capacity detection device for charge and discharge testing, the detection time is reduced by about 79.81%, and the detection energy consumption is reduced by about 79.78%; it can be seen that the model training and verification data involved in the present invention are all from the same batch of batteries, and the use of the method is not restricted by the types and models of batteries.

[0099] Example 3

[0100] In this example, in step S100, a batch of lithium iron phosphate square batteries of brand C is obtained, and the rated capacity C of the battery n is 40 Ah; in step S210, 5% of the batch of batteries to be detected are selected as conventional detection batteries; in steps S210 and S220, the constant current charging and discharging rates are both 0.6C, and the constant voltage charging cut-off current in step S210 is 0.05C; the charging time T in step S220 is 600S; the SOH Threshold in step S320 is 90%; the number k of resampling points in step S330 is 271; the remaining steps are the same as those in Example 1, and the same steps will not be repeated.

[0101] In this example, the classification model accurately classifies all the fast detection batteries according to the SOH Threshold of 90%, and the accuracy rate is 99.61%.

[0102] In this embodiment, the average absolute error of the fast detection battery SOH estimation that meets the requirements of cascade utilization is 1.57%, and the average relative error is 1.83%; compared with the charge and discharge test using the conventional capacity detection equipment, the detection time of this batch of batteries is reduced by about 94.63%, and the detection energy consumption is reduced by about 94.71%; it can be seen that the model training and verification data involved in the present invention all come from the same batch of batteries, and the use of the method is not limited by the types and models of batteries.

[0103] According to the disclosure and teaching of the above specification, those skilled in the art to which the present invention pertains can also make changes and modifications to the above embodiments. Therefore, the present invention is not limited to the specific embodiments disclosed and described above, and some modifications and changes to the present invention should also fall within the protection scope of the claims of the present invention. In addition, although some specific terms are used in this specification, these terms are only for convenience of description and do not constitute any limitation to the present invention.

Claims

1. A method for rapid detection of the health status of retired power lithium batteries, characterized in that: The following steps are involved: S100: Obtain a batch of retired battery cells of the same model and specification to be tested, and remove battery cells that are obviously not suitable for cascade utilization through appearance screening; S200: Using conventional capacity testing equipment to test the batteries to be tested and obtain data: Select 5-50% of the batteries in the batch of batteries to be tested as conventional test batteries, use conventional capacity testing equipment to perform a complete capacity calibration test and record the data; The remaining batteries are used as rapid test batteries, and regular capacity test equipment is used to perform rapid health status tests and record the data; S300: Processing the test data of the conventional test battery acquired in step S200: the test data of the conventional test battery is used for training and verification of the classification model and the SOH estimation model; S400: using the training verification data set generated after the data processing in step S300 to train the classification model and the SOH estimation model, to obtain the trained classification model and the SOH estimation model; S5 00: Process the test data of the rapid detection battery obtained in step S200: the test data of the rapid detection battery is used as a model input to achieve classification and SOH estimation of the rapid detection battery; S600: using the rapid detection battery data processed in step S500 to input into the classification model, determining whether the rapid detection battery meets the requirements of cascade utilization; S700: The battery data determined to meet the requirements of cascade utilization in step S600 is further input into the SOH estimation model to obtain an SOH estimation value, which is used as a rapid detection result of the health status of the battery.

2. A method for rapid detection of the health status of retired power lithium batteries according to claim 1, characterized in that: In S200, the battery to be tested is tested using conventional capacity testing equipment to obtain data, specifically including: S210: Randomly select 5-50% of the batch of batteries to be tested as regular test batteries, use regular capacity testing equipment, perform constant current discharge at a rate of 0.25C-1C, discharge to the lower limit of the battery voltage, let it stand for 10 minutes, perform a complete charge and discharge test and record the data; S220: The remaining batteries in the batch of batteries to be tested are used as rapid test batteries. Conventional capacity test equipment is used to perform constant current discharge at a rate of 0.25C-1C to the lower limit of the battery voltage. The batteries are left standing for 10 minutes, and a rapid health status test is performed and the data is recorded. The complete charge and discharge test includes: constant current charging at a rate of 0.25C-1C, charging to the upper limit of the battery voltage, then constant voltage charging to the cut-off current, standing for 10 minutes, and constant current discharging at a rate of 0.25C-1C, discharging to the lower limit of the battery voltage; The health status rapid detection test includes: constant current charging at a rate of 0.25C-1C, and the charging time T is 30 seconds-600 seconds; The recorded data include: the change of battery voltage, current and capacity over time during the test.

3. A method for rapid detection of the health status of retired power lithium batteries according to claim 1, characterized in that: In S300, the processing of the acquired conventional detection battery test data specifically includes: S310: Generate SOH estimation model training and verification data set labels, routinely detect the battery in the complete charge and discharge test, and the final discharge capacity is used as the battery's current discharge capacity C i , according to the capacity definition of battery health status SOH, the SOH of the conventional detection battery is calculated as follows: Among them C i is the current discharge capacity of the battery, C n is the rated capacity of the battery, and the calculated SOH is used as the label of the SOH estimation model training and validation dataset; S320: Generate classification model training and verification data set labels, and set the SOH threshold SOH to determine whether the retired battery meets the requirements of cascade utilization Threshold 70%-90%; when the SOH of the battery is tested routinely>SOH Threshold When the SOH of the battery meets the requirements of cascade utilization, the classification label is recorded as 1, indicating that the requirements are met; when the SOH of the battery is routinely tested <SOH Threshold When the SOH does not meet the requirements of cascade utilization, the classification label is recorded as 0, indicating that the requirements are not met; S330: Generate input features of the classification model and the SO estimation model training and verification data set, set the intercepted charging time T, and intercept the constant current charging stage data from the start of the test to the charging time T from the complete charging and discharging test data as the original data; U min is the minimum value of the voltage sequence in the intercepted data, U max is the maximum value of the voltage sequence in the intercepted data; set the number of resampling points k, based on the battery voltage upper limit U upper and the lower limit U lower As the boundary, generate uniformly distributed resampled voltage values; Where ΔU=0.005V; According to the original data, min ,U max ] range, and calculate the corresponding current, capacity and time values ​​by interpolation. lower ,U min ) and (U max ,U upper ] range, the current, capacity and time values ​​are all set to 0, and the voltage range [U lower ,U upper ] is a ternary sequence including current, capacity and time; the shape of the sequence is (k×3), which serves as the input feature of the classification model and the SOH estimation model training and verification data set; the input features correspond one-to-one to the labels of the classification model and the SOH estimation model, that is, each input feature sample corresponds to a classification label and a SOH estimation label; the above input features and their corresponding classification labels and SOH estimation labels are divided into a training set and a verification set to generate the classification model and the SOH estimation model training and verification data sets respectively, where the training set: verification set = 4:

1.

4. A method for rapid detection of the health status of retired power lithium batteries according to claim 1, characterized in that: In S400, the classification model and the SOH estimation model have similar structures, both of which are Inception-ResNet hybrid models based on causal dilated convolution, and the specific structure includes: an input layer, a mask layer, a multi-layer causal dilated convolution Inception-ResNet module and an output layer; The Inception-ResNet hybrid model combines the Inception module and the Residual Network (ResNet); The classification model and the SOH estimation model are basically the same in structure, with only the output layer being different. The output layer of the classification model uses the sigmoid activation function to obtain a probability value between 0 and 1, and then outputs 0 or 1 based on the classification threshold of 0.

5. The output layer of the SOH estimation model has no activation function and outputs a numerical value, namely the estimated value of SOH.

5. A method for rapid detection of the health status of retired power lithium batteries according to claim 1, characterized in that: In S500, the processing of the acquired rapid detection battery test data specifically includes: Generate input data for the classification model and the SOH estimation model, which are used to determine whether the SOH of the rapid detection battery to be detected meets the requirements of cascade utilization and estimate the SOH; use the constant current charging stage data in the health status rapid detection test data as the original data, and the constant current charging stage data includes voltage, current, capacity and time; U min is the minimum value of the voltage sequence in the original data, U max is the maximum value of the voltage sequence in the original data; set the number of resampling k, with the battery voltage upper limit U upper and the lower limit U lower As the boundary, generate uniformly distributed resampled voltage values; according to the original data, min ,U max ] range, and calculate the corresponding current, capacity and time values ​​by interpolation. lower ,U min ) and (U max ,U upper ] range, the current, capacity and time values ​​are all set to 0, and the voltage range [U lower ,U upper ] is a ternary sequence including current, capacity and time; the sequence shape is (k×3) and serves as the input of the trained classification model and SOH estimation model.

6. A method for rapid detection of the health status of retired power lithium batteries according to claim 1, characterized in that: In the S600, the rapid detection battery data after the data processing in step S500 is input into the classification model to determine whether the rapid detection battery meets the requirements of cascade utilization; wherein, the battery data determined to meet the requirements of cascade utilization continues with the subsequent SOH estimation process, and the battery data determined not to meet the requirements of cascade utilization no longer continues with the subsequent process and waits for scrapping and recycling.

Images