Similarity-Based SOH Prediction and SOC Estimation Methods for Lithium-Ion Batteries

Abstract

The present invention discloses a method for predicting the state of health (SOH) and estimating the state of charge (SOC) of a lithium-ion battery based on similarity. The method includes the following steps: obtaining the historical SOH sequence of lithium-ion batteries under the same type and working conditions, and performing preprocessing to train a first model; obtaining the data of the lithium-ion battery to be predicted; training a second model; predicting the SOH value of the lithium-ion battery to be predicted in the current charge-discharge half cycle; predicting the real-time SOC defined based on the rated capacity N and the SOC defined based on the maximum available capacity H ; updating the real-time SOC defined based on the rated capacity, the SOC defined based on the maximum available capacity N and the SOH in the current charge-discharge half cycle H ; updating the data to be predicted. The present invention provides a new research idea for the joint state estimation of lithium-ion batteries.

Description

Technical Field

[0001] The present invention relates to the field of battery state estimation, and particularly to a method for predicting the state of health (SOH) and estimating the state of charge (SOC) of a lithium-ion battery based on similarity. Background Art

[0002] In recent years, lithium-ion batteries have been recognized by the industry and supported by policies due to their high energy density, low self-discharge rate, high efficiency and stability. However, the estimation of their state of charge and state of health has always been a topic of concern for scientific researchers and users. Most of the existing studies focus on only one aspect, making it impossible to have a good grasp and management of the long-term and short-term scales of the battery.

[0003] The literature "A Method for Joint Estimation of SOC and SOH of Lithium Batteries Based on Dual Unscented Kalman Filters" realizes the joint estimation of SOC and SOH, but its accuracy highly depends on the equivalent circuit model, and it is necessary to obtain the identification values of the equivalent circuit parameters through experiments and use complex methods such as Kalman filtering to achieve the joint estimation of SOC and SOH. Summary of the Invention

[0004] To solve the above problems, the present invention proposes a method for predicting the state of health (SOH) and estimating the state of charge (SOC) of a lithium-ion battery based on similarity, which abandons complex models, does not require experiments, and has good accuracy.

[0005] The purpose of the present invention is achieved by at least one of the following technical solutions.

[0006] A method for predicting the state of health (SOH) and estimating the state of charge (SOC) of a lithium-ion battery based on similarity, comprising the following steps:

[0007] S1. Obtain the historical state of health (SOH) sequence of lithium-ion batteries under the same type and working conditions, and perform preprocessing;

[0008] S2. Train a first model according to the preprocessed historical state of health (SOH) sequence;

[0009] S3. Obtain the data of the lithium-ion battery to be predicted during the charge and discharge half-cycles, and divide it into training data and data to be predicted according to the charge and discharge half-cycles;

[0010] S4. Divide the training data into multiple training data samples according to a ratio in the order of time occurrence;

[0011] S5. Train a second model according to the training data samples;

[0012] S6. Use the trained first model to predict the state of health (SOH) value of the lithium-ion battery to be predicted during the current charge and discharge half-cycles;

[0013] S7. For the data to be predicted in this charge-discharge half-cycle of the lithium-ion battery, divide it into multiple data samples to be predicted according to the same ratio, and move each data sample to be predicted up and down as a whole in combination with the training data samples to obtain the translated data samples to be predicted. Input the translated data samples to be predicted into the trained second model, and predict the real-time state of charge (SOC) based on the rated capacity in this charge-discharge half-cycle of the lithium-ion battery to be predicted. N Obtain the predicted value, and based on the predicted value of the state of health (SOH) in this charge-discharge half-cycle in step S6, obtain the predicted value of the state of charge (SOC) based on the maximum available capacity in real time. H of the predicted value;

[0014] S8. Update the real-time state of charge (SOC) based on the rated capacity, N the state of charge (SOC) based on the maximum available capacity, H and the state of health (SOH) in this charge-discharge half-cycle, and respectively obtain the corresponding true values;

[0015] S9. Update the data to be predicted, and return to step S3.

[0016] Furthermore, in step S1, a complete charge-discharge cycle is completed when the lithium-ion battery completes a charging process and a discharging process successively. Completing a charging process or a discharging process alone is called a charge-discharge half-cycle;

[0017] The charging process includes a constant current charging stage or a constant voltage charging stage, and the discharging process includes a constant current discharging stage;

[0018] The ratio of the charging capacity or the discharging capacity to the rated capacity is called the state of health. Therefore, the rated capacity of the lithium-ion battery also needs to be given;

[0019] The state of health (SOH) sequence is composed of the states of health (SOH) of multiple charge-discharge half-cycles completed successively by the lithium-ion battery arranged in time sequence;

[0020] Preprocess the historical state of health (SOH) sequence as follows:

[0021] Divide the historical state of health (SOH) sequence into the states of health (SOH) in multiple charge-discharge half-cycles according to the charge-discharge half-cycles.

[0022] Furthermore, in step S2, given the window size, perform a sliding window on the historical state of health (SOH) sequence. Each time, take the states of health (SOH) in the window size of charge-discharge half-cycles as the input, and take the state of health (SOH) value in the 1 charge-discharge half-cycle after the window as the output, so as to establish the input and output of supervised learning. Adopt a relevant vector machine or an ensemble learning algorithm to train and obtain the first model;

[0023] The integrated learning algorithm includes random forest, extreme random tree, bagging algorithm, gradient boosting machine or decision tree.

[0024] Further, in step S3, an initial value p is given, where p > windows size;

[0025] Obtain the true values of the state of health SOH in the first p + 1 charge-discharge half-cycles completed successively by the lithium-ion battery to be predicted;

[0026] Obtain the data of the p-th charge-discharge half-cycle completed successively by the lithium-ion battery to be predicted as training data, and obtain the data of the (p + 2)-th charge-discharge half-cycle of the lithium-ion battery to be predicted as data to be predicted;

[0027] The training data includes the voltage, current, temperature and the state of charge SOC defined based on the rated capacity of the lithium-ion battery N curves for time;

[0028] The data to be predicted includes the curves of the voltage, current and temperature of the lithium-ion battery for time in the constant current charging stage, constant voltage charging stage or constant current discharging stage.

[0029] Further, in step S4, since it is difficult to ensure the accuracy of SOC estimation relying on the complete data curve, the data curve of the training data is divided into multiple training data samples according to a ratio in the order of time occurrence;

[0030] The physical quantities in each training data sample include voltage, current or temperature;

[0031] Record the initial values of the physical quantities corresponding to each training data sample, that is, the values of the physical quantities of the lithium-ion battery corresponding to the starting end of the data curve in each training data sample.

[0032] Further, in step S5, if the second model is trained using the relevance vector machine, new samples need to be obtained by interpolating the training data samples, and then the second model is trained according to the new samples; if the second model is trained using the integrated learning algorithm, no interpolation operation is required, and the second model is directly trained using the training data samples;

[0033] When training the second model using the relevance vector machine, step S5 includes the following steps:

[0034] S5.1. For the physical quantities in each training data sample, with time as the input and the physical quantity at the corresponding moment as the output, perform extreme value normalization operation to obtain the normalized training data samples, and train the extreme learning machine respectively;

[0035] S5.2. Set the number N of interpolation intervals. For the number N + 1 of interpolation points, divide the time in the training data samples after the normalization operation into N equal parts;

[0036] S5.3. Use the trained extreme learning machine to predict the physical quantities at N + 1 moments in the training data samples after the normalization operation;

[0037] S5.4. Denormalize the physical quantities at N + 1 moments in the training data samples after the normalization operation to restore them to the interpolated physical quantities, i.e., the new samples;

[0038] S5.5. Based on the new samples, use the corresponding physical quantities as inputs and the state of charge SOC N defined under the rated capacity as the output to train the second model;

[0039] When using the ensemble learning algorithm to train the second model, based on the training data samples, use the corresponding physical quantities as inputs and the state of charge SOC N defined under the rated capacity as the output to train the second model;

[0040] When training the second model, perform normalization processing on the inputs and outputs of the second model.

[0041] Further, in step S6, according to the true values of the state of health SOH in the p + 2 - window size - th to the p + 1 - th charge - discharge half - cycles of the lithium - ion battery to be predicted, input them into the first model and output the predicted value of the state of health SOH in the p + 2 - th charge - discharge half - cycle.

[0042] Further, in step S7, divide the data to be predicted in the p + 2 - th charge - discharge half - cycle of the lithium - ion battery into multiple data samples to be predicted according to the ratio in step S4 in the order of time occurrence; record the initial values of the physical quantities corresponding to each data sample to be predicted after division, subtract the initial value of the physical quantity corresponding to each data sample obtained in step S4 from the initial value of the physical quantity corresponding to each data sample to be predicted, and add the difference of the initial value of the physical quantity to each data sample to be predicted to form the translated data samples to be predicted;

[0043] Normalize each translated data sample to be predicted according to the normalization standard in step S6.5, and substitute the normalized translated data samples to be predicted into the trained second model, so as to obtain the predicted value of SOC N Perform denormalization on the predicted value to obtain the predicted value of the state of charge SOC N defined under the rated capacity in real - time, and then combine with step S6 to obtain the predicted value of SOC H of;

[0044] There are two ways to define the state of charge (SOC), which can be described by the ratio of the current available capacity to the rated capacity, i.e., the SOC based on the rated capacity definition. N It can also be described by the ratio of the current available capacity to the actual maximum charge-discharge capacity, i.e., the maximum available capacity, that is, the SOC based on the maximum available capacity definition. H The values of SOC under the two definitions are mutually converted through the state of health (SOH), specifically as follows:

[0045] SOCH = SOC N / SOH;

[0046] Combining the predicted value of the real-time SOC based on the rated capacity definition in the (p + 2)-th charge-discharge half-cycle N and the predicted value of the state of health SOH in the (p + 2)-th charge-discharge half-cycle, the predicted value of the real-time SOC based on the maximum available capacity definition in the (p + 2)-th charge-discharge half-cycle is obtained. H

[0047] Furthermore, in step S8, during the charge-discharge half-cycle, for the charging process, the formula SOC N = I * t / 3600 / rated capacity is used to determine the maximum value of the true value of SOC based on the rated capacity definition, where I is the current and t is the time length from the initial charging moment to the current moment; N For the discharging process, the formula SOC

[0048] = SOH - I * t / 3600 / rated capacity is used to determine the maximum value of the true value of SOC based on the rated capacity definition, and t is the time from the initial discharging moment to the current moment. N N

[0049] Furthermore, in step S9, p = p + 1, and return to step S3.

[0050] The present invention has the following advantages and effects compared with the prior art:

[0051] (1) There is no need to conduct an open-circuit test to measure data.

[0052] (2) There is no need for a complex modeling and parameter identification process, and it does not rely on an equivalent circuit model.

[0053] (3) For the first time, the relevance vector machine is applied to the joint estimation of the state of charge and the state of health of the battery, providing a new research idea for the joint state estimation of lithium-ion batteries. Description of the Drawings

[0054] To more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art.

[0055] Figure 1 It is the implementation flowchart in the embodiments of the present invention;

[0056] Figure 2 It is the voltage-time schematic diagram during a certain charging process of Cell1 in the embodiments of the present invention;

[0057] Figure 3 It is the temperature-time schematic diagram during a certain charging process of Cell1 in the embodiments of the present invention;

[0058] Figure 4 It is the voltage-time schematic diagram during a certain discharging process of Cell1 in the embodiments of the present invention;

[0059] Figure 5 It is the temperature-time schematic diagram during a certain charging process of Cell1 in the embodiments of the present invention. Specific implementation manners

[0060] The core of the present invention is to provide a method for predicting the state of health (SOH) and estimating the state of charge (SOC) of lithium-ion batteries based on similarity, making the joint state estimation of lithium-ion batteries simpler, more convenient, economical and feasible.

[0061] To enable those skilled in the art to better understand the technical solutions of the present invention, the following will further elaborate on the present invention in conjunction with the accompanying drawings and specific implementation manners. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0062] Embodiment 1:

[0063] A method for predicting the state of health (SOH) and estimating the state of charge (SOC) of lithium-ion batteries based on similarity, as Figure 1 shown, includes the following steps:

[0064] S1. Obtain the historical state of health (SOH) sequence of lithium-ion batteries under the same type and working conditions, and perform preprocessing;

[0065] A complete charge-discharge cycle of a lithium-ion battery is when it completes a charging process and a discharging process in sequence. Completing only a charging process or a discharging process alone is called a half charge-discharge cycle;

[0066] The charging process includes a constant current charging stage or a constant voltage charging stage, and the discharging process includes a constant current discharging stage;

[0067] The ratio of the charge capacity or discharge capacity to the rated capacity is called the state of health, so the rated capacity of the lithium-ion battery also needs to be given;

[0068] The state-of-health SOH sequence is formed by arranging the state-of-health SOHs of multiple charge-discharge half-cycles completed by the lithium-ion battery in sequence;

[0069] Preprocess the historical state-of-health SOH sequence as follows:

[0070] Divide the historical state-of-health SOH sequence into the state-of-health SOHs in multiple charge-discharge half-cycles according to the charge-discharge half-cycles.

[0071] In this embodiment, it is carried out under the Oxford battery public dataset. The Oxford battery public dataset includes a total of 8 batteries such as Cell1 to Cell8. The voltage, temperature curve, rated capacity (0.74 Ah), and state-of-charge SOC annotation value (obtaining the electric quantity by integrating the current, and then dividing the electric quantity by (3600×0.74) to obtain the state-of-charge SOC defined based on the rated capacity) of each discharge process are known. In addition, the state-of-health SOHs of the charge-discharge processes of Cell1, Cell2, Cell5, and Cell6 are also known (the state-of-health SOHs of the charge-discharge processes of Cell1, Cell2, Cell5, and Cell6 are used for training).

[0072] S2. Train the first model according to the preprocessed historical state-of-health SOH sequence;

[0073] In this embodiment, given the window size windows size = 2, perform sliding window on the historical state-of-health SOH sequence. Each time, take the state-of-health SOH values in windows size charge-discharge half-cycles as the input, and take the state-of-health SOH value in 1 charge-discharge half-cycle after the window as the output, so as to establish the input and output of supervised learning. Use the relevant vector machine or ensemble learning algorithm to train and obtain the first model;

[0074] The ensemble learning algorithm includes random forest, extremely randomized trees, bagging algorithm, gradient boosting machine, or decision tree.

[0075] S3. Obtain the data of the lithium-ion battery to be predicted in the charge-discharge half-cycle, and divide it into training data and data to be predicted according to the charge-discharge half-cycle;

[0076] Given the initial value p, p > windows size;

[0077] Obtain the true values of the state-of-health SOHs in the first p + 1 charge-discharge half-cycles completed by the lithium-ion battery to be predicted;

[0078] Obtain the data of the p-th charge-discharge half-cycle completed by the lithium-ion battery to be predicted as training data, and obtain the data of the (p + 2)-th charge-discharge half-cycle of the lithium-ion battery to be predicted as the data to be predicted;

[0079] The training data includes the voltage, current, temperature, and state of charge SOC defined based on the rated capacity of the lithium-ion battery N versus time curves;

[0080] The data to be predicted includes the curves of voltage, current, and temperature of the lithium-ion battery versus time in the constant current charging stage, constant voltage charging stage, or constant current discharging stage.

[0081] In this embodiment, obtain the data of the lithium-ion battery Cell3 during the charge-discharge half-cycle.

[0082] S4. Divide the training data into multiple training data samples according to the time occurrence order and proportion;

[0083] Since it is difficult to ensure the accuracy of SOC estimation relying on the complete data curve, divide the data curve of the training data into multiple training data samples according to the time occurrence order and proportion;

[0084] The physical quantities in each training data sample include voltage, current, or temperature;

[0085] Record the initial values of the physical quantities corresponding to each training data sample, that is, the values of the physical quantities of the lithium-ion battery corresponding to the starting end of the data curve in each training data sample.

[0086] In this embodiment, record the initial temperature value corresponding to each training data sample, that is, the temperature of the lithium-ion battery corresponding to the starting end of the data curve in each training data sample.

[0087] From Figure 2 、 Figure 3 、 Figure 4 and Figure 5 it can be seen that the curves of two consecutive charges or two consecutive discharges are extremely similar. At the same time, the two similar curves have certain differences. The vertical dotted lines in the figure divide the charge curve and the discharge curve into different parts. Among them, for charging, it is divided into four parts, denoted as the first charging sample ch01, the second charging sample ch02, the third charging sample ch03, and the fourth charging sample ch04, with a ratio of 0.05:0.3:0.35:0.3; for discharging, it is divided into three parts, denoted as the first discharge sample dis01, the second discharge sample dis02, and the third discharge sample dis03, with a ratio of 0.05:0.9:0.05;

[0088] In addition, record the corresponding initial temperature values in the first charging sample ch01, the second charging sample ch02, the third charging sample ch03, the fourth charging sample ch04, the first discharging sample dis01, the second discharging sample dis02, and the third discharging sample dis03, including the initial temperature value T01 of the first charging sample, the initial temperature value T02 of the second charging sample, the initial temperature value T03 of the third charging sample, the initial temperature value T04 of the fourth charging sample, the initial temperature value T05 of the first discharging sample, the initial temperature value T06 of the second discharging sample, and the initial temperature value T07 of the third discharging sample.

[0089] S5. Train a second model based on the training data samples;

[0090] If a relevant vector machine is used to train the second model, interpolation is required for the training data samples to obtain new samples, and then the second model is trained based on the new samples; if an ensemble learning algorithm is used to train the second model, no interpolation operation is required, and the second model is directly trained using the training data samples;

[0091] In this embodiment, uniform interpolation is performed on the voltage curves and temperature curves of the first charging sample ch01, the second charging sample ch02, the third charging sample ch03, the fourth charging sample ch04, the first discharging sample dis01, the second discharging sample dis02, and the third discharging sample dis03 to obtain corresponding new samples, including the first charging new sample ch11, the second charging new sample ch12, the third charging new sample ch13, the fourth charging new sample ch14, the first discharging new sample dis11, the second discharging new sample dis12, and the third discharging new sample dis13;

[0092] In this embodiment, taking the first charging new sample ch11 as an example, when using a relevant vector machine to train the second model, step S5 includes the following steps:

[0093] S5.1. For the physical quantities in the first charging sample ch01, use time as the input and the physical quantities at the corresponding moments as the output, perform extreme value normalization operation to obtain the first charging sample ch01 after the normalization operation, and train an extreme learning machine using the gradient descent method respectively;

[0094] S5.2. Set the number of interpolation interval N = 100. For the number of interpolation points N + 1, divide the time in the first charging sample ch01 after the normalization operation into N equal parts;

[0095] S5.3. Use the trained extreme learning machine to predict the voltage or temperature at N + 1 moments in the first charging sample ch01 after the normalization operation;

[0096] S5.4. Antide-normalize the physical quantities at N + 1 moments in the first charging sample ch01 after the normalization operation to restore them to the interpolated voltage or temperature, i.e., the first charging new sample ch11;

[0097] S5.5. In this embodiment, taking the first charging new sample ch11 as an example, on the basis of the new sample, using the voltage or temperature corresponding to the first charging new sample ch11 as the input, and based on the state of charge SOC N as the output, train the second model;

[0098] When training the second model using the ensemble learning algorithm, on the basis of the training data samples, using the corresponding physical quantities as the input, and based on the state of charge SOC N as the output, train the second model;

[0099] When training the second model, perform normalization processing on the input and output of the second model.

[0100] S6. Use the trained first model to predict the state of health SOH value of the lithium-ion battery in the current charge-discharge half-cycle to be predicted;

[0101] According to the true values of the state of health SOH in the p + 2 - windows size to p + 1 charge-discharge half-cycles of the lithium-ion battery to be predicted, input them into the first model, and output the predicted value of the state of health SOH in the p + 2nd charge-discharge half-cycle.

[0102] S7. For the data to be predicted in the current charge-discharge half-cycle of the lithium-ion battery, divide it into multiple data samples to be predicted according to the same ratio, and combine the training data samples to move each data sample to be predicted up and down as a whole to obtain the translated data samples to be predicted. Input them into the trained second model to predict the state of charge SOC N predicted value based on the rated capacity definition in the current charge-discharge half-cycle of the lithium-ion battery to be predicted, and obtain the predicted value of the state of charge SOC H predicted value based on the maximum available capacity definition in real time according to the predicted value of the state of health SOH in the current charge-discharge half-cycle in step S6;

[0103] In this embodiment, divide the data to be predicted in the p + 2nd charge-discharge half-cycle of the lithium-ion battery Cell3 into multiple data samples to be predicted according to the ratio in step S4 in the order of time occurrence, including the first data sample to be predicted for charging ch21, the second data sample to be predicted for charging ch22, the third data sample to be predicted for charging ch23, and the fourth data sample to be predicted for charging ch24, as well as the first data sample to be predicted for discharging dis21, the second data sample to be predicted for discharging dis22, and the third data sample to be predicted for discharging dis23;

[0104] Record the initial values of the physical quantities corresponding to each data sample to be predicted after division, denoted as the initial value of the temperature of the first charging sample to be predicted T21, the initial value of the temperature of the second charging sample to be predicted T22, the initial value of the temperature of the third charging sample to be predicted T23, the initial value of the temperature of the fourth charging sample to be predicted T24, the initial value of the temperature of the first discharging sample to be predicted T25, the initial value of the temperature of the second discharging sample to be predicted T26, and the initial value of the temperature of the third discharging sample to be predicted T27;

[0105] Subtract the initial value of the physical quantity corresponding to each data sample obtained in step S4 from the initial value of the physical quantity corresponding to each data sample to be predicted, i.e., T21 - T01, T22 - T02, T23 - T03, T24 - T04, T25 - T05, T26 - T06, T27 - T07;

[0106] Add the difference in the initial values of the physical quantity to each data sample to be predicted to form the translated data samples to be predicted, including the first charging translated sample ch31, the second charging translated sample ch32, the third charging translated sample ch33, and the fourth charging translated sample ch34, as well as the first discharging translated sample dis31, the second discharging translated sample dis32, and the third discharging translated sample dis33;

[0107] Normalize each translated data sample to be predicted according to the normalization standard in step S6.5, and substitute the normalized translated data sample to be predicted into the trained second model to obtain the predicted value of SOC N of the predicted value, and perform inverse normalization on the predicted value to obtain the real-time state of charge SOC defined based on the rated capacity N of the predicted value, and then combine step S6 to obtain the predicted value of SOC H of the predicted value;

[0108] There are two definition methods for the state of charge SOC (state of charge). It can be described by the ratio of the current available capacity to the rated capacity, that is, the state of charge SOC defined based on the rated capacity N , or it can be described by the ratio of the current available capacity to the actual maximum charge and discharge capacity, that is, the maximum available capacity, that is, the state of charge SOC defined based on the maximum available capacity H , and the values of the state of charge SOC under the two definitions are mutually converted through the state of health SOH, specifically as follows:

[0109] SOC H＝ SOC N / SOH;

[0110] Combined with the predicted value of the state of charge (SOC) based on the rated capacity in the real-time of the (p + 2)-th charge and discharge half-cycle and the predicted value of the state of health (SOH) in the (p + 2)-th charge and discharge half-cycle, the predicted value of the state of charge (SOC) based on the maximum available capacity in the real-time of the (p + 2)-th charge and discharge half-cycle is obtained. N For the predicted value of the state of charge (SOC) based on the rated capacity in the real-time of the current charge and discharge half-cycle, and the predicted value of the state of health (SOH) in the current charge and discharge half-cycle, the true values of the corresponding states of charge (SOC) based on the maximum available capacity in the real-time of the current charge and discharge half-cycle are obtained. H of the predicted value.

[0111] S8. For the state of charge (SOC) based on the rated capacity in the real-time of the current charge and discharge half-cycle N , the state of charge (SOC) based on the maximum available capacity H and the state of health (SOH) are updated to obtain the corresponding true values respectively;

[0112] During the charge and discharge half-cycle, for the charging process, the formula SOC N = I * t / 3600 / rated capacity is used to determine the maximum value of the true value of the SOC N based on the rated capacity, where I is the current and t is the time length from the initial charging moment to the current moment;

[0113] For the discharging process, the formula SOC N = SOH - I * t / 3600 / rated capacity is used to determine the maximum value of the true value of the SOC N based on the rated capacity, and t is the time from the initial discharging moment to the current moment.

[0114] S9. p = p + 1, and return to step S3.

[0115] The prediction results of steps S5 and S6 are quantitatively evaluated using the mean absolute error (MAE) and the maximum error (ME). x i and x i are respectively the true value and the predicted value of the real-time SOC or SOH in the i-th half-cycle sampling times:

[0116]

[0117]

[0118] K is the number of half-cycle samplings.

[0119] Example 2:

[0120] The difference between this example and Example 1 is that in step S5, the data to be predicted of the lithium-ion battery Cell4 in the current charge and discharge half-cycle is obtained.

[0121] Example 3:

[0122] The difference between this embodiment and Embodiment 1 is that in step S5, the data to be predicted of the lithium-ion battery Cell7 in the current charge-discharge half-cycle is obtained.

[0123] Embodiment 4:

[0124] The difference between this embodiment and Embodiment 1 is that in step S5, the data to be predicted of the lithium-ion battery Cell8 in the current charge-discharge half-cycle is obtained.

[0125] Combined with Embodiment 1, Embodiment 2, Embodiment 3 and Embodiment 4, Table 1 makes a quantitative evaluation of the prediction results of S5.

[0126] Table 1 Charge-discharge SOC prediction errors of different cycles of Cell3, Cell4, Cell7 and Cell8

[0127]

[0128]

[0129] Table 2 makes a quantitative evaluation of the prediction results of S6.

[0130] Table 2 SOH prediction errors of Cell3, Cell4, Cell7 and Cell8

[0131] Cell3 Cell4 Cell7 Cell8 MAE 8.13209E-06 1.62596E-07 3.11916E-07 1.59981E-07 ME 0.0301 0.0078 0.0092 0.0069

[0132] For the real-time SOC defined based on the maximum available capacity H , a new index, i.e., the qualification rate, is introduced. It is defined that when the prediction error of the real-time SOC H defined based on the maximum available capacity does not exceed 5%, the prediction is regarded as qualified. Thus, for each charge-discharge half-cycle, its qualification rate is mostly higher than 97%, which means that for all sampling points, the method adopted in the present invention can ensure that more than 97% of the predictions are reliable.

[0133] The above embodiments are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are all included in the protection scope of the present invention.

Claims

1. Similarity-based SOH Prediction and SOC Estimation Method for Lithium-ion Batteries Characterized in that It includes the following steps: S1. Obtain the historical state of health (SOH) sequence of lithium-ion batteries under the same type and working conditions, and perform preprocessing; S2. Train the first model according to the preprocessed historical SOH sequence; S3. Obtain the data of the lithium-ion battery to be predicted during the charge and discharge half-cycles, and divide it into training data and data to be predicted according to the charge and discharge half-cycles; S4. Divide the training data into multiple training data samples according to a certain proportion in the order of time occurrence; S5. Train the second model according to the training data samples; If the second model is trained using the relevance vector machine, new samples need to be obtained by interpolating the training data samples, and then the second model is trained according to the new samples; if the second model is trained using the ensemble learning algorithm, no interpolation operation is required, and the second model is directly trained using the training data samples; When the second model is trained using the relevance vector machine, step S5 includes the following steps: S5.

1. For each physical quantity in the training data sample, use time as the input and the physical quantity at the corresponding moment as the output, and perform extreme value normalization operation to obtain the normalized training data sample, and train the extreme learning machine respectively; S5.

2. Set the number of interpolation intervals N. For the number of interpolation points N + 1, divide the time in the normalized training data sample into N equal parts; S5.

3. Use the trained extreme learning machine to predict the physical quantities at N + 1 moments in the normalized training data sample; S5.

4. Denormalize the physical quantities at N + 1 moments in the normalized training data sample to restore them to the interpolated physical quantities, that is, new samples; S5.

5. On the basis of the new samples, using the corresponding physical quantity as the input and the state of charge SOC defined based on the rated capacity as the output, train the second model; N As the output, train the second model; When training the second model using the ensemble learning algorithm, based on the training data samples, with the corresponding physical quantity as the input and the state of charge (SOC) defined under the rated capacity as the output, train the second model; N ​ When training the second model, normalize the input and output of the second model; S6. Use the trained first model to predict the SOH value of the state of health of the lithium-ion battery to be predicted during the current charge and discharge half-cycle; S7. For the data to be predicted in this charge-discharge half-cycle of the lithium-ion battery, divide it into multiple data samples to be predicted according to the same ratio, and move each data sample to be predicted up and down as a whole in combination with the training data samples to obtain the translated data samples to be predicted. Input the translated data samples to be predicted into the trained second model, and predict the real-time state of charge (SOC) based on the rated capacity in this charge-discharge half-cycle of the lithium-ion battery to be predicted. N Obtain the predicted value of SOC, and obtain the predicted value of SOC based on the maximum available capacity according to the predicted value of the state of health (SOH) in this charge-discharge half-cycle in step S6. H The predicted value; S8. Update the state of charge SOC defined based on the rated capacity, the state of charge SOC defined based on the maximum available capacity, and the state of health SOH in real time during this charge-discharge half-cycle, and obtain the corresponding true values respectively; N The state of charge SOC defined based on the maximum available capacity H and update the state of health SOH to obtain the corresponding true values respectively; S9. Update the data to be predicted, and return to step S3.

2. The similarity-based SOH prediction and SOC estimation method for lithium-ion batteries according to claim 1 Characterized in that In step S1, a complete charge and discharge cycle of a lithium-ion battery is completed by first performing a charging process and then a discharging process. Completing a charging process or a discharging process alone is called a charge and discharge half-cycle; The charging process includes a constant current charging stage or a constant voltage charging stage, and the discharging process includes a constant current discharging stage; The state of health (SOH) sequence is composed of the SOH of the state of health of multiple charge and discharge half-cycles completed by the lithium-ion battery arranged in time sequence; The preprocessing of the historical SOH sequence is as follows: Divide the historical SOH sequence into the SOH of the state of health in multiple charge and discharge half-cycles according to the charge and discharge half-cycles.

3. The similarity-based SOH prediction and SOC estimation method for lithium-ion batteries according to claim 2 Characterized in that In step S2, given the window size, a sliding window is applied to the historical state of health (SOH) sequence. Each time, the SOH values of windows size charge-discharge half-cycles are taken as inputs, and the SOH value of the 1 charge-discharge half-cycle after the window is taken as the output, thereby establishing the input and output of supervised learning. Using a relevance vector machine or an ensemble learning algorithm, a first model is trained. The ensemble learning algorithm includes random forest, extremely randomized trees, bagging algorithm, gradient boosting machine, or decision tree.

4. The similarity-based lithium-ion battery SOH prediction and SOC estimation method according to claim 3, characterized in that in step S3, an initial value p is given, where p > windows size; obtain the true values of the state of health (SOH) of the lithium-ion battery to be predicted in the first p + 1 charge-discharge half-cycles completed successively; obtain the data of the p-th charge-discharge half-cycle completed successively by the lithium-ion battery to be predicted as training data, and obtain the data of the (p + 2)-th charge-discharge half-cycle of the lithium-ion battery to be predicted as the data to be predicted; The training data includes the voltage, current, temperature, and state of charge (SOC) defined based on the rated capacity of a lithium-ion battery during the constant current charging stage, constant voltage charging stage, or constant current discharging stage. N The curve for time; The data to be predicted includes the curves of voltage, current, and temperature with respect to time during the constant current charging stage, constant voltage charging stage, or constant current discharging stage of the lithium-ion battery.

5. The similarity-based lithium-ion battery SOH prediction and SOC estimation method according to claim 4, characterized in that in step S4, the data curves of the training data are divided into multiple training data samples according to a ratio in the order of time occurrence; The physical quantities in each training data sample include voltage, current, or temperature; Record the initial values of the physical quantities corresponding to each training data sample, that is, the values of the physical quantities of the lithium-ion battery corresponding to the starting end of the data curve in each training data sample.

6. The similarity-based lithium-ion battery SOH prediction and SOC estimation method according to claim 1, characterized in that in step S6, according to the true values of the state of health (SOH) of the lithium-ion battery to be predicted in the (p + 2 - windows size)-th to (p + 1)-th charge-discharge half-cycles, input them into the first model, and output the predicted value of the state of health (SOH) of the (p + 2)-th charge-discharge half-cycle.

7. The similarity-based lithium-ion battery SOH prediction and SOC estimation method according to claim 6, characterized in that in step S7, divide the data to be predicted in the (p + 2)-th charge-discharge half-cycle of the lithium-ion battery into multiple data samples to be predicted according to the ratio in step S4 in the order of time occurrence; record the initial values of the physical quantities corresponding to each divided data sample to be predicted, subtract the initial value of the physical quantity corresponding to each data sample obtained in step S4 from the initial value of the physical quantity corresponding to each data sample to be predicted, and add the difference of the initial values of the physical quantities to each data sample to be predicted correspondingly to form the translated data samples to be predicted; Normalize each translated data sample to be predicted according to the normalization standard in step S6.5, and substitute the normalized translated data sample to be predicted into the trained second model to obtain the predicted value of SOC N Then, denormalize the predicted value to obtain the real-time state of charge SOC defined based on the rated capacity N of the predicted value, and then combine with step S6 to obtain the predicted value of SOC H of the predicted value; There are two ways to define the state of charge (SOC). It can be described by the ratio of the current available capacity to the rated capacity, that is, the SOC based on the rated capacity definition. N Or it can be described by the ratio of the current available capacity to the actual maximum charge-discharge capacity, that is, the maximum available capacity, that is, the SOC based on the maximum available capacity definition. H The values of the SOC under the two definitions are mutually converted through the state of health (SOH), as follows: SOC H＝ SOC N / SOH; Combined with the predicted value of the state of charge SOC defined based on the rated capacity in the (p + 2)-th charge-discharge half-cycle and the predicted value of the state of health SOH in the (p + 2)-th charge-discharge half-cycle, the predicted value of the state of charge SOC defined based on the maximum available capacity in the (p + 2)-th charge-discharge half-cycle is obtained. N H ​ 8. The similarity-based lithium-ion battery SOH prediction and SOC estimation method according to claim 7, characterized in that In step S8, during the charge-discharge half cycle, for the charging process, the formula SOC N = I * t / 3600 / rated capacity is used to determine the maximum value of the true value of SOC N defined based on the rated capacity, where I is the current and t is the time length from the initial charging moment to the current moment; For the discharge process, the formula SOC N = SOH - I * t / 3600 / rated capacity is used to determine the maximum value of the true value of SOC N defined based on the rated capacity, where t is the time from the initial discharge moment to the current moment.

9. The similarity-based SOH prediction and SOC estimation method for lithium-ion batteries according to any one of claims 4 to 8, characterized in that, in step S9, p = p + 1, and return to step S3.

Images