A on-site battery health assessment method based on multi-impedance measurement

Abstract

The present invention discloses a method for on-site battery health assessment based on multi-impedance measurement. The present invention includes the following steps: measuring the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies of an on-site battery under any on-site working conditions; predicting the real part impedance values and imaginary part impedance values corresponding to multiple identical preset frequencies under laboratory working conditions; predicting the electrochemical impedance spectrum under laboratory working conditions; predicting the charge / discharge curve and relaxation voltage curve under laboratory working conditions; performing feature extraction and health assessment to obtain the health assessment result of the on-site battery. The present invention uses machine learning to bridge on-site data and laboratory data, and compared with the existing battery health assessment methods based on on-site historical data, it has the advantages of high precision, low cost, fast speed, being usable anytime and anywhere, and not involving privacy data, etc.

Description

Technical Field

[0001] The present invention belongs to an online health assessment method in the field of batteries, and particularly relates to a field battery health assessment method based on multi-impedance measurement. Background Art

[0002] Batteries have achieved large-scale commercial applications in portable electronic devices, transportation vehicles, and stationary energy storage fields. There is an urgent need for longer cycle life and safer batteries in these fields. Precise health assessment (including health diagnosis and health prognosis) information of batteries can provide a reliable basis for predictive maintenance and management strategy optimization of on-site batteries, thereby reducing the occurrence of safety accidents and extending the service life of batteries. In recent years, thanks to the rapid development of artificial intelligence technology and the open source of numerous battery test data sets, data-driven methods combining machine learning and feature engineering have achieved excellent performance in battery health assessment and demonstrated great application potential in all aspects of the battery's entire life cycle, such as research and development, production, research, application, and recycling. However, most current data-driven methods and their feature engineering rely on laboratory data at specific stages under specific working conditions, such as charge and discharge data, impedance data, relaxation data, etc. Different from laboratory data with relatively stable operating environments and working conditions, on-site battery data is affected by the coupling of many random and uncertain factors such as application scenarios, operating environments, and user habits, and has great randomness and uncertainty. Therefore, the vast majority of data-driven methods developed based on laboratory data cannot be directly applied to on-site data.

[0003] Researchers have also conducted some research on battery health assessment methods based on field data. The main idea is to develop data-driven methods based on the massive historical data generated by field battery operation. According to the field battery's needs for operation, maintenance and optimization management, the ideal battery health assessment method should have five main characteristics: (1) High accuracy. The higher the accuracy of the health assessment method, the more accurate the predictive maintenance and management strategy optimization of the field battery can be, which is more beneficial to the safe and long-life operation of the field lithium battery. On the contrary, inaccurate health assessment may have a negative impact on the safe and long-life operation of the field battery. (2) Low cost. The battery health assessment method should have low development cost, low deployment cost, and low use cost, which is more conducive to large-scale commercial promotion. (3) Use anytime and anywhere. The battery health assessment method can carry out field battery health assessment anytime and anywhere according to user needs. (4) Fast. The battery health assessment method can quickly complete the field battery health assessment and will not have a negative impact on the normal operation of the field battery. (5) Does not involve user privacy data. As users' awareness of data privacy continues to increase, the protection of battery operation data will become more and more stringent, and the amount of data available will become less and less. The data on which the battery health assessment method relies should try not to involve the user's privacy data, such as historical operation data, which may include historical travel data, historical charging data, and user usage habit data.

[0004] Compared with the five characteristics that an ideal battery health assessment method should have, the existing data-driven methods developed based on massive field historical data are not ideal in all aspects. In terms of accuracy, under the same data volume, the health assessment accuracy of the field data-based method is not yet as good as that of the laboratory data-based method. In terms of cost, the field data-based method needs to accumulate massive field historical data, and the development cost is high. In terms of use, the field data-based method needs to collect complete or specific stage operation data, which cannot be used anytime and anywhere, and the speed is also slow. In terms of privacy, the field data-based method uses massive historical data, which may involve the user's privacy data. In addition, the existing field data-based health assessment methods have not made full use of the many excellent research results based on laboratory data. Most of the methods are developed from scratch, which is a waste of research resources. According to the first principle idea, based on the excellent results of many laboratory data-based methods, the development of the lowest cost field battery health assessment method should make use of the existing excellent results as much as possible, rather than developing from scratch. Summary of the invention

[0005] To solve the problem of on-site battery health assessment, the present invention proposes an on-site battery health assessment method based on multi-impedance measurement. The present invention adopts a brand-new idea different from the existing methods: in actual on-site applications, by online measuring the impedances of a small number of on-site batteries and combining with a machine learning model prediction, high-quality laboratory data of the batteries under laboratory conditions can be obtained. Since there have been a large number of research results on battery health assessment based on high-quality laboratory data and excellent performance has been achieved, the existing numerous research results can be directly utilized for high-precision battery health assessment.

[0006] The solution adopted by the present invention is as follows:

[0007] I. An on-site battery health assessment method based on multi-impedance measurement

[0008] 1) Online measure the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies of on-site batteries under any on-site working conditions;

[0009] 2) According to the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the measured on-site working conditions, respectively predict the real part impedance values and imaginary part impedance values corresponding to the same multiple preset frequencies under laboratory conditions;

[0010] 3) According to the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the predicted laboratory conditions, predict the EIS data under laboratory conditions;

[0011] 4) Input the EIS data under laboratory conditions into the laboratory condition charge Q / V data prediction model, laboratory condition discharge Q / V data prediction model, and laboratory condition relaxation V / t data prediction model respectively, and respectively predict the charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory conditions;

[0012] 5) Conduct feature extraction and battery health assessment based on the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory conditions to obtain the health assessment result of the on-site battery.

[0013] In the above step 2), the prediction process for the real part impedance value and imaginary part impedance value corresponding to any preset frequency under the measured on-site working conditions is specifically as follows:

[0014] Input the real part impedance value corresponding to the current preset frequency under the on-site working conditions into the real part impedance value prediction model of the preset frequency under laboratory conditions to obtain the real part impedance value corresponding to the preset frequency under laboratory conditions; input the imaginary part impedance value corresponding to the current preset frequency under the on-site working conditions into the imaginary part impedance value prediction model of the preset frequency under laboratory conditions to obtain the imaginary part impedance value corresponding to the preset frequency under laboratory conditions.

[0015] The specific content of 3) is as follows:

[0016] Input the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the laboratory conditions obtained by prediction into the laboratory condition EIS prediction model, and predict the EIS data under the laboratory conditions.

[0017] In 5), the health assessment result of the on-site battery includes the aging state and / or the remaining service life.

[0018] When the health assessment result of the on-site battery is the aging state, extract diagnostic features based on the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under the laboratory conditions, and input the extracted diagnostic features into the battery health diagnosis model, and the model outputs the current aging state of the on-site battery.

[0019] When the health assessment result of the on-site battery is the remaining service life, extract prognostic features based on the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under the laboratory conditions, and input the extracted prognostic features into the battery health prognosis model, and the model outputs the current remaining service life of the on-site battery.

[0020] The multiple preset frequencies are obtained by selection through the expert experience method, filtering method, wrapper method, embedding method, or clustering method.

[0021] II. An on-site battery health assessment system based on multi-impedance measurement

[0022] An impedance data acquisition unit for the on-site battery, which is used to acquire the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies of the on-site battery under any on-site conditions;

[0023] An impedance data generation unit for the laboratory conditions, which is used to generate the real part impedance values and imaginary part impedance values corresponding to multiple identical preset frequencies under the laboratory conditions according to the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the measured on-site conditions;

[0024] An EIS data generation unit for the laboratory conditions, which is used to predict the EIS data under the laboratory conditions according to the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the predicted laboratory conditions;

[0025] A charge Q / V data, discharge Q / V data, and relaxation V / t data generation unit for the laboratory conditions, which is used to predict the charge Q / V data, discharge Q / V data, and relaxation V / t data under the laboratory conditions according to the EIS data under the predicted laboratory conditions;

[0026] A health assessment unit, which is used to perform feature extraction based on the predicted EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory conditions, so as to obtain the health assessment result of the on-site battery.

[0027] III. A computer device

[0028] The device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the method are implemented.

[0029] IV. A computer-readable storage medium

[0030] The medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the method are implemented.

[0031] The beneficial effects of the present invention are as follows:

[0032] 1. The present invention proposes a method for on-site battery health assessment based on multi-impedance measurement. Compared with the existing methods based on on-site historical data, it has the advantages of high precision, low cost, fast speed, being able to be used anytime and anywhere, and not involving privacy data, etc.;

[0033] 2. The present invention uses machine learning to bridge the on-site data and laboratory data of the battery, filling the gap between the research results of battery health assessment based on laboratory data and on-site applications, so that all research results of battery health assessment based on laboratory data can be directly applied to actual on-site batteries;

[0034] 3. The present invention uses the battery aging state index and remaining service life index under laboratory conditions to accurately quantify and evaluate the current health state of the on-site battery. Compared with the features / indicators extracted from random on-site battery data, the battery health assessment result is more accurate.

[0035] 4. The present invention subverts the existing fixed mode of battery health assessment based on on-site data, and opens up a new direction for the online health assessment of on-site batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is the overall flowchart of the present invention.

[0037] Figure 2 is a schematic diagram of the real part impedance value and imaginary part impedance value corresponding to two preset frequencies (f1 = 10.0Hz and f2 = 312.5Hz) under on-site conditions and laboratory conditions in an embodiment of the present invention.

[0038] Figure 3 is a comparison diagram of the EIS prediction data and actual measurement data under laboratory conditions in an embodiment of the present invention.

[0039] Figure 4 It is a comparison graph of the relaxation time distribution (DRT) data corresponding to the EIS obtained by prediction and the DRT data corresponding to the EIS obtained by actual measurement in the embodiments of the present invention.

[0040] Figure 5 It is a comparison graph of the predicted values and the actual measured values of the charging Q / V data and the discharging Q / V data under the laboratory conditions in the embodiments of the present invention.

[0041] Figure 6 It is a comparison graph of the incremental capacity (IC) data corresponding to the charging Q / V data obtained by prediction and the IC data corresponding to the charging Q / V data obtained by actual measurement in the embodiments of the present invention.

[0042] Figure 7 It is a comparison graph of the differential voltage (DV) data corresponding to the charging Q / V data obtained by prediction and the DV data corresponding to the charging Q / V data obtained by actual measurement in the embodiments of the present invention. Detailed implementation manners

[0043] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] In this embodiment, the selected battery chemistry type is NCM, the manufacturer is LG, the model is INR18650HG2, and the rated capacity is 3000 mAh. The number of preset frequencies is set to two. The two preset frequencies are specifically a certain value in the range of 50 mHz to 10 KHz. f1 is named the first frequency and f2 is named the second frequency.

[0045] The two preset frequencies (f1 and f2) can be selected by the expert experience method, the filtering method, the wrapper method, the embedding method or the clustering method. In this embodiment, the K-Means clustering algorithm in the clustering method is used to select the two preset frequencies. Specifically, 16 data points in the high-frequency part (1 Hz to 1 KHz) of the EIS are selected, and each data point includes a frequency value, a real part impedance value and an imaginary part impedance value. The 16 data points are divided into two clusters, and the K-means clustering algorithm is used to extract two preset frequencies from the 16 frequencies corresponding to the 16 data points. Among them, the number of clusters of the K-means clustering algorithm is set to 2. The 16 data points are input into the K-means clustering algorithm for clustering to obtain the cluster labels of each data point. After clustering, calculate the mean value of the frequencies corresponding to all data points in each cluster. Next, within each cluster, find the data point closest to the mean frequency within the cluster, and use the frequency value corresponding to this data point as the preset frequency of the cluster. The two obtained preset frequency values are f1 = 10.0 Hz and f2 = 312.5 Hz.

[0046] Such as Figure 1As shown in the figure, the present invention includes the following steps:

[0047] 1) Online measure the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies of the on-site battery under any on-site working conditions, as Figure 2 shown. Any on-site working condition includes any value within the battery operating temperature range and any value within the battery SOC range. In this embodiment, the battery operating temperature range is [0°C, 40°C], and the battery SOC range is [0%, 100%].

[0048] 2) According to the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the measured on-site working conditions, respectively predict the real part impedance values and imaginary part impedance values corresponding to the same multiple preset frequencies under laboratory working conditions. The laboratory working conditions include a preset battery operating temperature and a preset battery SOC. In this embodiment, the battery operating temperature under laboratory working conditions is 25°C, and the battery SOC is 90% during the charging process. Input the real part impedance value corresponding to the first frequency f1 under on-site working conditions into the real part impedance value prediction model of the laboratory working conditions at the first frequency, and obtain the real part impedance value corresponding to the first frequency f1 under laboratory working conditions. Input the imaginary part impedance value corresponding to the first frequency f1 under on-site working conditions into the imaginary part impedance value prediction model of the laboratory working conditions at the first frequency f1, and obtain the imaginary part impedance value corresponding to the first frequency f1 under laboratory working conditions. Use the same method to predict the real part impedance value and imaginary part impedance value corresponding to the second frequency under laboratory working conditions. The input-output data range for training the laboratory working condition impedance value prediction models for each preset frequency should cover the entire battery life cycle, the entire operating temperature range, and the entire SOC range. In this embodiment, the two laboratory working condition impedance value prediction models use a random forest (RF) regression model, the battery operating temperature range is [0°C, 40°C], the battery SOC range is [0%, 100%], and the entire battery life cycle is that the retained capacity under laboratory working conditions decays from the initial 100% to 80%. The multiple preset frequencies described in step 2) are exactly the same as the multiple preset frequencies in step 1).

[0049] 3) Input the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the predicted laboratory working conditions into the laboratory working condition EIS prediction model together, and predict the EIS data under laboratory working conditions, as Figure 3 shown. Perform relaxation time distribution processing on the EIS data predicted under laboratory working conditions and the measured EIS data respectively and convert them into DRT curves. Comparing the errors between the two DRT curves can further verify the accuracy of the EIS data prediction, as Figure 4As shown. The input and output data ranges for training the EIS prediction model under laboratory conditions should cover the entire life cycle of the battery. In this embodiment, the RF regression model is used for the EIS prediction model under laboratory conditions. The input of the EIS prediction model under laboratory conditions is the real part impedance values ( and ) and the imaginary part impedance values ( and ) at two frequencies (f1 and f2) under laboratory conditions, and the output is the EIS prediction data (1 Hz to 1 KHz) under laboratory conditions.

[0050] 4) Input the EIS data under laboratory conditions into the prediction models for the charging capacity vs. voltage (Q / V) data, the discharging Q / V data, and the relaxation voltage vs. time (V / t) data under laboratory conditions respectively, and predict the charging Q / V data, the discharging Q / V data, and the relaxation V / t data under laboratory conditions respectively, as shown in Figure 5 (a) of Figure 5 and Figure 6 . Take the derivative of the capacity of the predicted charging Q / V data and the measured charging Q / V data under laboratory conditions and convert them into the incremental charging capacity (IC) curve respectively. Comparing the error between the two IC curves can further verify the accuracy of the prediction of the charging Q / V data, as shown in Figure 7As shown. By using the same method, the accuracy of the prediction of the discharge Q / V data and the relaxation V / t data can be verified. The inputs of the laboratory condition charge Q / V data prediction model, the laboratory condition discharge Q / V data prediction model, and the laboratory condition relaxation V / t data prediction model are all the EIS data obtained by prediction under laboratory conditions. The output of the laboratory condition charge Q / V data prediction model is the constant current charge Q / V data or the constant voltage charge capacity vs. current (Q / I) data under laboratory conditions. The output of the laboratory condition discharge Q / V data prediction model is the constant current discharge Q / V data or the constant voltage discharge Q / I data under laboratory conditions. The output of the laboratory condition relaxation V / t data prediction model is the relaxation V / t data (the relaxation V / t data after charging or the relaxation V / t data after discharging) under laboratory conditions. The input and output data ranges used for training the laboratory condition charge Q / V data prediction model, the laboratory condition discharge Q / V data prediction model, and the laboratory condition relaxation V / t data prediction model should cover the entire life cycle of the battery. In this embodiment, the constant current charge current under laboratory conditions is 1 / 3C, the constant current charge cut-off voltage is 4.2V, the constant voltage charge voltage is 4.2V, and the constant voltage charge cut-off current is 1 / 20C. The constant current discharge current under laboratory conditions is 1 / 3C, the constant current discharge cut-off voltage is 2.5V, the constant voltage discharge voltage is 2.5V, and the constant voltage discharge cut-off current is 1 / 20C. The laboratory condition charge Q / V data prediction model, the laboratory condition discharge Q / V data prediction model, and the laboratory condition relaxation V / t data prediction model all use the RF regression model.

[0051] 5) Feature extraction and battery health assessment are performed based on the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory conditions to obtain the health assessment results of the on-site battery. The features extracted from the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory conditions include instantaneous features, statistical features, model parameter features, etc. The health assessment results of the on-site battery include the aging state and / or the remaining service life. When the health assessment result of the on-site battery is the aging state, diagnostic features are extracted from the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory conditions, and the extracted diagnostic features are input into the battery health diagnosis model. The model outputs the current aging state of the on-site battery, such as the retained capacity, internal resistance, lithium inventory loss, active material loss, etc. of the current battery under laboratory conditions. When the health assessment result of the on-site battery is the remaining service life, prognostic features are extracted from the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory conditions, and the extracted prognostic features are input into the battery health prognosis model. The model outputs the current remaining service life of the on-site battery, such as the remaining number of cycles, remaining cumulative charge and discharge capacity, etc. of the battery under laboratory conditions. The input and output data ranges for training the battery health diagnosis model and the battery health prognosis model should cover the entire life cycle of the battery. In this embodiment, the features extracted from the EIS data, charge Q / V data, and discharge Q / V data are statistical features. Both the battery health diagnosis model and the battery health prognosis model use the RF regression model. The battery aging state index uses the retained capacity of the battery under laboratory conditions, and the battery remaining service life index uses the remaining number of cycles of the battery under laboratory conditions.

[0052] Finally, it should be noted that the above embodiments and descriptions are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the disclosure of the technical solutions of the present invention, and they should all be covered by the protection scope of the claims of the present invention.

Claims

1. A method for on-site battery health assessment based on multi-impedance measurement, characterized in that, Including the following steps: 1) Online measure the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies of the on-site battery under any on-site working conditions; 2) According to the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the measured on-site working conditions, respectively predict the real part impedance values and imaginary part impedance values corresponding to the same multiple preset frequencies under laboratory working conditions; 3) According to the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the predicted laboratory working conditions, predict the EIS data under laboratory working conditions; 4) Input the EIS data under laboratory working conditions into the laboratory working condition charge Q / V data prediction model, the laboratory working condition discharge Q / V data prediction model, and the laboratory working condition relaxation V / t data prediction model respectively, and respectively predict the charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory working conditions; 5) Perform feature extraction and battery health assessment based on the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory working conditions to obtain the health assessment result of the on-site battery.

2. The on-site battery health assessment method based on multi-impedance measurement according to claim 1, wherein In step 2), the prediction process of the real part impedance value and the imaginary part impedance value corresponding to any preset frequency under the measured on-site working conditions is specifically as follows: Input the real part impedance value corresponding to the current preset frequency under the on-site working conditions into the real part impedance value prediction model of the laboratory working conditions for this preset frequency to obtain the real part impedance value corresponding to this preset frequency under laboratory working conditions; Input the imaginary part impedance value corresponding to the current preset frequency under the on-site working conditions into the imaginary part impedance value prediction model of the laboratory working conditions for this preset frequency to obtain the imaginary part impedance value corresponding to this preset frequency under laboratory working conditions.

3. The on-site battery health assessment method based on multi-impedance measurement according to claim 1, wherein, Step 3) is specifically as follows: Input the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies under the predicted laboratory working conditions into the laboratory working condition EIS prediction model together to predict the EIS data under laboratory working conditions.

4. The on-site battery health assessment method based on multi-impedance measurement according to claim 1, characterized in that In step 5), the health assessment result of the on-site battery includes the aging state and / or the remaining service life.

5. The on-site battery health assessment method based on multi-impedance measurement according to claim 4, characterized in that When the health assessment result of the on-site battery is the aging state, extract diagnostic features based on the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory working conditions, and input the extracted diagnostic features into the battery health diagnosis model, and the model outputs the current aging state of the on-site battery.

6. The on-site battery health assessment method based on multi-impedance measurement according to claim 4, characterized in that When the health assessment result of the on-site battery is the remaining service life, extract prognostic features based on the EIS data, charge Q / V data, discharge Q / V data, and relaxation V / t data under laboratory working conditions, and input the extracted prognostic features into the battery health prognosis model, and the model outputs the current remaining service life of the on-site battery.

7. The on-site battery health assessment method based on multi-impedance measurement according to claim 1, wherein The multiple preset frequencies are obtained by the expert experience method, the filtering method, the wrapper method, the embedding method, or the clustering method.

8. A on-site battery health assessment system based on multi-impedance measurement, characterized in that Including: The impedance data acquisition unit of the on-site battery, which is used to acquire the real part impedance values and imaginary part impedance values corresponding to multiple preset frequencies of the on-site battery under any on-site working conditions; The impedance data generation unit under laboratory conditions is used to generate the real - part impedance values and imaginary - part impedance values corresponding to multiple preset frequencies under laboratory conditions based on the real - part impedance values and imaginary - part impedance values corresponding to multiple preset frequencies under on - site conditions obtained by measurement; The EIS data generation unit under laboratory conditions is used to predict the EIS data under laboratory conditions based on the real - part impedance values and imaginary - part impedance values corresponding to multiple preset frequencies under laboratory conditions obtained by prediction; The charging Q / V data, discharging Q / V data, and relaxation V / t data generation unit under laboratory conditions is used to predict the charging Q / V data, discharging Q / V data, and relaxation V / t data under laboratory conditions based on the EIS data under laboratory conditions obtained by prediction; The health assessment unit is used to perform feature extraction based on the EIS data, charging Q / V data, discharging Q / V data, and relaxation V / t data under laboratory conditions obtained by prediction, and obtain the health assessment result of the on - site battery.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images