A semi-supervised SOH estimation method for lithium-ion batteries based on charging fragments

Abstract

The present invention proposes a semi-supervised SOH estimation method for lithium-ion batteries based on charging fragments, comprising: collecting lithium-ion battery operating data to construct an original data set; calculating the SOH value of each cycle based on charge and discharge integrity; extracting health features from the original data set based on desirability and correlation to construct a semi-supervised total sample set; dividing the semi-supervised total sample set into a labeled sample set and an unlabeled sample set; establishing a self-training recurrent neural network model to generate pseudo labels and update the labeled sample set; iteratively training the self-training recurrent neural network model until the unlabeled sample set is empty or reaches a preset number of iterations; and using the final labeled sample set to establish a lithium-ion battery SOH model based on the self-training recurrent neural network model. The present invention can solve the problems of fragmented charging and missing SOH labels faced in lithium-ion battery SOH estimation, making the SOH estimation method more applicable to actual application scenarios and improving the practicality and reliability of SOH estimation.

Description

Technical Field

[0001] The present invention belongs to the field of lithium-ion battery SOH estimation, and in particular relates to a semi-supervised SOH estimation method for lithium-ion batteries based on charging segments. Background Art

[0002] Lithium-ion batteries have been widely used in recent years in fields such as electric vehicles, microelectronic devices, and energy storage power stations due to their long lifespan, low self-discharge rate, and high energy density. However, under the influence of factors such as charge and discharge depth, current rate, and external environment, lithium-ion batteries inevitably produce some side reactions. These side reactions gradually erode the battery's state of health (SOH), causing its performance to decline. When the SOH of a lithium-ion battery degrades below the safety threshold, the probability of safety accidents such as fire and explosion increases significantly. Therefore, accurately estimating the SOH of a lithium-ion battery is particularly critical to ensuring its safe operation.

[0003] Extensive research has been conducted on SOH estimation for lithium-ion batteries, but most of this research has been limited to lithium-ion batteries that have undergone a complete charge-discharge cycle. In the diverse application scenarios of lithium-ion batteries, they often encounter challenges such as discontinuous and fragmented charge-discharge cycles and the lack of SOH labels for lithium-ion batteries. The existence of fragmented charge-discharge phenomena makes it impossible to directly extract health assessment indicators from complete charge-discharge curves, thereby increasing the complexity of feature engineering. At the same time, the lack of a large number of SOH labels has caused SOH estimation to shift from a supervised learning problem to a semi-supervised problem with partial SOH labels, increasing the difficulty of data-driven model training.

[0004] Therefore, it is urgent to propose a flexible SOH estimation method for scenarios where lithium-ion batteries are fragmented in charge and discharge and SOH labels are missing. Summary of the Invention

[0005] The purpose of the present invention is to provide a semi-supervised SOH estimation method for lithium-ion batteries based on charging fragments to overcome the defects of existing methods and realize SOH estimation of lithium-ion batteries in the case of fragmented charging and discharging and missing SOH labels.

[0006] The purpose of the present invention can be achieved by the following technical solution: a semi-supervised SOH estimation method for lithium-ion batteries based on charging fragments, the method comprising the following steps:

[0007] Step S1: Collecting time series data of voltage, current and temperature during the operation of the lithium-ion battery to construct a lithium-ion battery original data set;

[0008] Step S2: Each charging event of the lithium-ion battery is recorded as a cycle, and several discharge events after each charging event belong to the cycle until the next charging event occurs; the original data set of the lithium-ion battery is traversed, and the SOH value of each cycle of the lithium-ion battery is calculated based on the charge and discharge integrity;

[0009] Step S3: extract health features from the lithium-ion battery original dataset based on desirability and relevance, and form a semi-supervised total sample set together with the SOH value calculated in step S2;

[0010] Step S4: Divide the total semi-supervised sample set in step S3 into a labeled sample set containing SOH values and an unlabeled sample set not containing SOH values;

[0011] Step S5: Using the labeled sample set to establish a self-training recurrent neural network model, the unlabeled sample set is input into the self-training recurrent neural network model to generate SOH pseudo-labels corresponding to each cycle in the unlabeled sample set, and the SOH pseudo-labels that meet the confidence requirements and the corresponding health features in the unlabeled sample set are added to the labeled sample set and deleted from the unlabeled sample set;

[0012] Step S6: Repeat step S5 until the unlabeled sample set is empty or the preset maximum number of iterations is reached, and output the labeled sample set after the final iteration of the self-training recurrent neural network model;

[0013] Step S7: Using the labeled sample set after the final iteration of the self-training recurrent neural network model in step S6, a model with the same structure and parameters as the self-training recurrent neural network model in step S5 is trained to establish a lithium-ion battery SOH estimation model;

[0014] Step S8: Collect data from the actual operation of the lithium-ion battery, extract the same health characteristics as in step S3, and input them into the lithium-ion battery SOH estimation model in step S7 in the same data format to obtain the SOH estimation value of the current cycle lithium-ion battery.

[0015] Furthermore, step S2 specifically includes the following steps:

[0016] S201: traverse the lithium-ion battery original data set, and determine the charging event and discharging event of each cycle according to the charge and discharge protocol, and whether the lithium-ion battery has experienced a complete charge or a complete discharge;

[0017] S202: Based on the judgment result in step S201, the SOH value at the loop is calculated according to the following rules:

[0018] If at cycle i, the lithium-ion battery has undergone a complete discharge, the calculation formula for SOH at cycle i is:

[0019]

[0020] If at cycle i, the lithium-ion battery has only undergone a complete charge and not a complete discharge, the calculation formula for SOH at cycle i is:

[0021]

[0022] If at cycle i, the lithium-ion battery has not undergone a complete charge and a complete discharge, then the SOH at cycle i is empty;

[0023] Among them, t dc1 ,t dc2 The discharge start time and discharge end time are respectively, ch1 ,t ch2 They are the charging start time and the charging end time, I CH (t) is the charging current, I DC (t) is the discharge current, C rated is the rated capacity of the lithium-ion battery.

[0024] Furthermore, in step S3, data preprocessing is performed on the original data set of lithium-ion batteries, outliers in the data are removed, and missing values are filled by interpolation. After the data preprocessing, health features are extracted.

[0025] Furthermore, step S3 specifically includes the following steps:

[0026] S301: Analyze the correlation between the voltage time difference of the main health characteristics, specifically: in all complete charging cycles, the maximum value of the starting voltage of the traversal charging voltage curve Minimum cut-off voltage Any two voltage points V m and V n Corresponding timestamp and Calculate the equal voltage time difference The correlation analysis method was used to calculate the different V m and V n The equal voltage time difference Δt f The correlation with the SOH of lithium-ion batteries in a complete charging cycle is sorted in descending order of correlation to obtain a decreasing sequence of correlations of equal voltage time differences;

[0027] S302: Analyze the desirability of the main health characteristics such as voltage time difference in charging events, specifically: Statistical calculation of the charging start voltage of each cycle in the original data set of lithium-ion batteries Charge cut-off voltage Calculate the average charge start voltage and average charge cut-off voltage Traverse the decreasing sequence of equal voltage time difference correlation in step S301, and the first one in the sequence is included in The optimal equal voltage time difference within the range is considered as the optimal equal voltage time difference considering both relevance and desirability. Right now The correlation is the highest across all cycles, covering the most charging events;

[0028] S303: Extracting the charging start voltage in each cycle of the lithium-ion battery charging segment Charge cut-off voltage Optimal equal voltage time difference Charging start temperature Charging end temperature As health features, constitute the lithium-ion battery health feature set

[0029] S304: Align the lithium-ion battery health feature set with the SOH value calculated in step S2 in a cyclic order to establish a semi-supervised total sample set S i =[F i ,SOH i ].

[0030] Furthermore, step S5 specifically includes the following steps:

[0031] S501: There will be a labeled sample set The self-training recurrent neural network model is trained using the training set processed by the sliding window algorithm. The structure and hyperparameters of the self-training recurrent neural network model are adjusted by observing the decrease in the loss of the training set and the validation set. The training of the self-training recurrent neural network model is completed when the loss of both the training set and the validation set no longer decreases.

[0032] S502: Unlabeled sample set After being processed by the sliding window algorithm, the self-training recurrent neural network model trained in step S501 is input to output the pseudo labels corresponding to each cycle in the unlabeled sample set.

[0033] S503: Confidence judgment is performed on the pseudo labels generated by the self-training recurrent neural network model on the unlabeled sample set. If the SOH pseudo labels of the cycle If it is within the 95% SOH confidence interval of the two labeled cycles before and after it, it is considered that the pseudo label generated by the model in this cycle meets the error requirements;

[0034] S504: Add the SOH pseudo labels that meet the confidence requirements and the corresponding health features to the labeled sample set, and delete them from the unlabeled sample set.

[0035] Furthermore, the calculation formula of the 95% SOH confidence interval is as follows:

[0036]

[0037] in, and are the mean and standard deviation of the SOH in the closed intervals of the two labeled cycles f1 and f2 before and after the pseudo label; n is the total number of battery cycles in the closed intervals of the two labeled cycles before and after. When the significance level α = 0.05, the confidence level of the interval is 95%. According to the table method, the Z score Z α / 2 About 1.96.

[0038] Furthermore, the SOH is linearly interpolated between the two labeled cycles f1 and f2 before and after the pseudo label, and the mean and standard deviation of the interpolated curve are used instead. and

[0039] Furthermore, the self-training recurrent neural network in step S5 includes: a self-training simple RNN neural network, a self-training LSTM neural network, and a self-training GRU neural network.

[0040] Furthermore, the self-training recurrent neural network adopts a self-training GRU neural network, which includes a GRU layer with 256 neurons, a GRU layer with 128 neurons, a fully connected layer with 32 neurons, and a fully connected layer with 1 neuron.

[0041] The present invention has the following beneficial effects:

[0042] This invention fully considers issues such as fragmented charging and missing SOH labels that exist in the actual use of lithium-ion batteries, making the estimation method more realistic and practical, improving the practicality and reliability of SOH estimation. Specifically, through in-depth analysis of the desirability and relevance of lithium-ion battery health characteristics, health characteristics that can cover most fragmented charging events are extracted; by combining a semi-supervised self-training method with a recurrent neural network model, unlabeled data is effectively utilized, significantly improving the utilization rate of data resources and reducing the reliance on complete SOH label data. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of a semi-supervised SOH estimation method for lithium-ion batteries based on charging segments provided by the present invention;

[0044] Figure 2 It is the self-training recurrent neural network model structure in the embodiment of the present invention;

[0045] Figure 3is the estimation result of the semi-supervised SOH estimation method for lithium-ion batteries based on charging segments in an embodiment of the present invention;

[0046] Figure 4 2. It is a schematic diagram of correlation analysis of voltage-time differences of main health characteristics in an embodiment of the present invention;

[0047] Figure 5 3 is a schematic diagram of performing confidence judgment on pseudo labels generated by a model in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0049] like Figure 1 As shown, this embodiment provides a semi-supervised SOH estimation method for lithium-ion batteries based on charging segments, comprising the following steps:

[0050] Step S1: Collecting time series data of voltage, current and temperature during the operation of the lithium-ion battery to construct a lithium-ion battery original data set;

[0051] This example uses four 18650 lithium-ion batteries with a rated capacity of 2Ah as examples. Three of these batteries are used for model training, and one is used for model testing. Cyclic charge-discharge experiments are performed on the lithium-ion batteries with randomized charge-discharge starting voltages and charge-discharge depths until the maximum discharge capacity of the lithium-ion batteries falls below 70% of the nominal capacity. Time, current, voltage, and temperature data are collected during the charge and discharge process.

[0052] Step S2: Each charging event of the lithium-ion battery is recorded as a cycle, and several discharge events after each charging event belong to the cycle until the next charging event occurs; the original data set of the lithium-ion battery is traversed, and the SOH value of each cycle of the lithium-ion battery is calculated based on the charge and discharge integrity.

[0053] Step S3: Extract health features from the lithium-ion battery original dataset based on desirability and relevance, and form a semi-supervised total sample set together with the SOH value calculated in step S2.

[0054] Step S4: Divide the total semi-supervised sample set in step S3 into a labeled sample set containing SOH values and an unlabeled sample set not containing SOH values.

[0055] Step S5: Using the labeled sample set to establish a self-training recurrent neural network model, the unlabeled sample set is input into the self-training recurrent neural network model to generate SOH pseudo-labels corresponding to each cycle in the unlabeled sample set, and the SOH pseudo-labels that meet the confidence requirements and the corresponding health features in the unlabeled sample set are added to the labeled sample set and deleted from the unlabeled sample set;

[0056] The self-training GRU neural network model structure and parameters used in this embodiment are as follows: Figure 2 As shown, it contains a GRU layer with 256 neurons, a GRU layer with 128 neurons, a fully connected layer with 32 neurons, and a fully connected layer with 1 neuron.

[0057] Step S6: Repeat step S5 until the unlabeled sample set is empty or reaches the preset maximum number of iterations, and output the labeled sample set after the final iteration of the self-training recurrent neural network model.

[0058] Step S7: Using the labeled sample set after the final iteration of the self-training recurrent neural network model in step S6, a model with the same structure and parameters as the self-training recurrent neural network model in step S5 is trained to establish a lithium-ion battery SOH estimation model.

[0059] Step S8: Collect data from the actual operation of the lithium-ion battery, extract the same health characteristics as in step S3, and input them into the lithium-ion battery SOH estimation model in step S7 in the same data format to obtain the SOH estimation value of the current cycle lithium-ion battery.

[0060] The estimation results of the semi-supervised SOH estimation method for lithium-ion batteries based on charging segments provided in this embodiment are as follows: Figure 3 shown.

[0061] Furthermore, step S2 specifically includes the following steps:

[0062] S201: traverse the lithium-ion battery original data set, and determine the charging event and discharging event of each cycle according to the charge and discharge protocol, and whether the lithium-ion battery has experienced a complete charge or a complete discharge;

[0063] S202: Based on the judgment result in step S201, the SOH value at the loop is calculated according to the following rules:

[0064] If at cycle i, the lithium-ion battery has undergone a complete discharge, the calculation formula for SOH at cycle i is:

[0065]

[0066] If at cycle i, the lithium-ion battery has only undergone a complete charge and not a complete discharge, the calculation formula for SOH at cycle i is:

[0067]

[0068] If at cycle i, the lithium-ion battery has not undergone a complete charge and a complete discharge, then the SOH at cycle i is empty;

[0069] Among them, t dc1 ,t dc2 The discharge start time and discharge end time are respectively, ch1 ,t ch2 They are the charging start time and the charging end time, I CH (t) is the charging current, I DC (t) is the discharge current, C rated is the rated capacity of the lithium-ion battery.

[0070] Furthermore, step S3 specifically includes the following steps:

[0071] S301: Preprocessing the original data set of lithium-ion batteries, removing outliers in the data, and filling missing values using interpolation;

[0072] S302: Analyze the correlation between the voltage time difference and the main health characteristics;

[0073] In all complete charging cycles, the maximum voltage at the beginning of the charging voltage curve is reached. Minimum cut-off voltage Any two voltage points V m and V n Corresponding timestamp and Calculate the equal voltage time difference The Pearson correlation analysis method was used to calculate the different V m and V n The correlation between the equal voltage time difference and the SOH of lithium-ion batteries in a complete charging cycle is sorted in descending order of correlation to obtain the correlation between different V m and V n The decreasing sequence of the correlation between the equal voltage time difference and SOH; where f is the fth complete charging cycle of the lithium-ion battery.

[0074] like Figure 4 Shown are the charging voltage curves for all complete charging cycles of a data set. The cycle number in the color bar on the right is the number of complete charging cycles in the total number of cycles (i.e., cycle 2, cycle 54, cycle 106, ..., cycle 1092, and cycle 1144 are complete charging cycles in the total number of cycles).

[0075] In all complete charge cycles, any V m and Calculate the V m and V n Corresponding timestamp and Calculate the equal voltage time difference Then, the Pearson correlation analysis method was used to calculate the different V m and V n The obtained Δt f Correlation between α and SOH.

[0076] For example, V m and V n If the battery is charged from 4.05V to 4.15V, the time required to charge the battery from 4.05V to 4.15V is calculated in all complete charging cycles to form a sequence [Δt2, Δt 54 ,Δt 106 ,...,Δt f ,...,Δt 1144 ], calculate the difference between this sequence and SOH([SOH2,SOH 54 ,SOH 106 ,...,SOH f ,...,SOH 1144 ]), this correlation is recorded as Then in and Take other V m and V n Calculate the correlation Finally, we get a V m and V n List of values and SOH correlations Then sort the list from most to least relevant.

[0077] S303: Further analyzing the desirability of the main health characteristics such as voltage time difference in charging events;

[0078] Count the charging start voltage of each cycle in the original data set of lithium-ion batteries Charge cut-off voltage Calculate the average charge start voltage and average charge cut-off voltage Wherein, N is the total number of cycles; traverse the decreasing sequence of equal voltage time difference correlation in step S302, and the first one in the sequence is included in The optimal equal voltage time difference within the range is considered as the optimal equal voltage time difference considering both relevance and desirability. Right now The correlation is the highest across all cycles, covering the most charging events;

[0079] S304: Extracting the charging start voltage in each cycle of the lithium-ion battery charging segment Charge cut-off voltage Optimal equal voltage time difference Charging start temperature Charging end temperature As health features, constitute the lithium-ion battery health feature set

[0080] S305: Align the lithium-ion battery health feature set with the SOH value calculated in step S2 in a cyclic order to establish a semi-supervised total sample set S i =[F i ,SOH i ]; where i is the i-th cycle of the lithium-ion battery.

[0081] Furthermore, step S5 specifically includes the following steps:

[0082] S501: There will be a labeled sample set The data is divided into a training set and a validation set in a ratio of 8:2. The self-training recurrent neural network model is trained using the training set processed by the sliding window algorithm. The structure and hyperparameters of the self-training recurrent neural network model are adjusted by observing the decrease in the loss of the training set and the validation set. The training of the self-training recurrent neural network model is completed when the loss of both the training set and the validation set no longer decreases.

[0083] S502: Unlabeled sample set After being processed by the sliding window algorithm, the self-training recurrent neural network model trained in step S501 is input to output the pseudo labels corresponding to each cycle in the unlabeled sample set.

[0084] S503: Confidence judgment is performed on the pseudo labels generated by the self-training recurrent neural network model on the unlabeled sample set. If the SOH pseudo labels of the cycle If the value is within the 95% SOH confidence interval of the two labeled cycles before and after it, then the pseudo labels generated by the model in this cycle are considered to meet the error requirements. The 95% SOH confidence interval calculation formula is as follows:

[0085]

[0086] in, and are the mean and standard deviation of SOH in the closed intervals of the two labeled cycles f1 and f2 before and after the pseudo label, respectively, and are replaced by the mean and standard deviation of the linear interpolation of SOH in this interval; n = f2-f1+1 is the total number of battery cycles in the closed interval of the two labeled cycles before and after. When the significance level α = 0.05, the confidence level of this interval is 95%. According to the table method, the Z score Z at this time α / 2 About 1.96;

[0087] When the pseudo labels generated by the self-training recurrent neural network model on cycle i Within the 95% confidence interval of the two labeled cycles SOH before and after it (such as Figure 5 The pseudo labels generated by the model meet the requirements when the value of the interpolated curve is (green area in the middle). Where i∈(f1,f2), n is the number of cycles between f1 and f2. The interpolation of the mean and standard deviation means linearly interpolating the SOH between f1 and f2, calculating the mean and standard deviation of the interpolated curve, and replacing and

[0088] S504: Add the SOH pseudo labels that meet the confidence requirements and the corresponding health features to the labeled sample set, and delete them from the unlabeled sample set.

[0089] In addition, the self-training recurrent neural network can also adopt a self-training simple RNN neural network or a self-training LSTM neural network, but is not limited thereto.

[0090] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A semi-supervised SOH estimation method for lithium-ion batteries based on charging fragments, characterized in that: include: Step S1: Collecting time series data of voltage, current and temperature during the operation of the lithium-ion battery to construct a lithium-ion battery original data set; Step S2: Each charging event of the lithium-ion battery is recorded as a cycle, and several discharge events after each charging event belong to the cycle until the next charging event occurs; the original data set of the lithium-ion battery is traversed, and the SOH value of each cycle of the lithium-ion battery is calculated based on the charge and discharge integrity; Step S3: extracting health features from the original lithium-ion battery dataset based on desirability and relevance, and forming a semi-supervised total sample set together with the SOH value calculated in step S2; specifically, the following steps are included: S301: Analyze the correlation between the voltage time difference of the main health characteristics, specifically: in all complete charging cycles, the maximum value of the starting voltage of the traversal charging voltage curve Minimum cut-off voltage Any two voltage points V m and V n Corresponding timestamp and Calculate the equal voltage time difference The correlation analysis method was used to calculate the different V m and V n The equal voltage time difference Δt f The correlation with the SOH of lithium-ion batteries in a complete charging cycle is sorted in descending order of correlation to obtain a decreasing sequence of correlations of equal voltage time differences; S302: Analyze the desirability of the main health characteristics such as voltage time difference in charging events, specifically: Statistically calculate the charging start voltage V of each cycle in the original data set of the lithium-ion battery i S , charging cut-off voltage V i E , calculate the average charge starting voltage and average charge cut-off voltage Traverse the decreasing sequence of equal voltage time difference correlation in step S301, and the first one in the sequence is included in The optimal equal voltage time difference within the range is considered as the optimal equal voltage time difference considering both relevance and desirability. Right now The correlation is the highest across all cycles, covering the most charging events; S303: Extracting the charging start voltage V in each cycle of the lithium-ion battery charging segment i S , charging cut-off voltage V i E , optimal equal voltage time difference Charging start temperature T i S , charging end temperature T i E As health features, constitute the lithium-ion battery health feature set S304: Align the lithium-ion battery health feature set with the SOH value calculated in step S2 in a cyclic order to establish a semi-supervised total sample set S i =[F i ,SOH i ]; Step S4: Divide the total semi-supervised sample set in step S3 into a labeled sample set containing SOH values and an unlabeled sample set not containing SOH values; Step S5: Using the labeled sample set to establish a self-training recurrent neural network model, the unlabeled sample set is input into the self-training recurrent neural network model, and the SOH pseudo-label corresponding to each cycle in the unlabeled sample set is generated. The SOH pseudo-labels that meet the confidence requirements and the corresponding health features in the unlabeled sample set are added to the labeled sample set and deleted from the unlabeled sample set. The specific steps include: S501: There will be a labeled sample set The self-training recurrent neural network model is trained using the training set processed by the sliding window algorithm. The structure and hyperparameters of the self-training recurrent neural network model are adjusted by observing the decrease in the loss of the training set and the validation set. The training of the self-training recurrent neural network model is completed when the loss of both the training set and the validation set no longer decreases. S502: Unlabeled sample set After being processed by the sliding window algorithm, the self-training recurrent neural network model trained in step S501 is input to output the pseudo labels corresponding to each cycle in the unlabeled sample set. S503: Confidence judgment is performed on the pseudo labels generated by the self-training recurrent neural network model on the unlabeled sample set. If the SOH pseudo labels of the cycle If the value lies within the 95% SOH confidence interval of the two labeled cycles before and after it, the pseudo-label generated by the model in this cycle is considered to meet the error requirements; the calculation formula of the 95% SOH confidence interval is as follows: in, and are the mean and standard deviation of the SOH in the closed intervals of the two labeled cycles f1 and f2 before and after the pseudo label; n is the total number of battery cycles in the closed intervals of the two labeled cycles before and after. When the significance level α = 0.05, the confidence level of the interval is 95%. According to the table method, the Z score Z α / 2 About 1.96; S504: adding the SOH pseudo labels that meet the confidence requirements and the corresponding health features to the labeled sample set, and deleting them from the unlabeled sample set; Step S6: Repeat step S5 until the unlabeled sample set is empty or the preset maximum number of iterations is reached, and output the labeled sample set after the final iteration of the self-training recurrent neural network model; Step S7: Using the labeled sample set after the final iteration of the self-training recurrent neural network model in step S6, a model with the same structure and parameters as the self-training recurrent neural network model in step S5 is trained to establish a lithium-ion battery SOH estimation model; Step S8: Collect data from the actual operation of the lithium-ion battery, extract the same health characteristics as in step S3, and input them into the lithium-ion battery SOH estimation model in step S7 in the same data format to obtain the SOH estimation value of the current cycle lithium-ion battery.

2. The semi-supervised SOH estimation method for lithium-ion batteries based on charging segments according to claim 1 is characterized in that: Step S2 specifically includes the following steps: S201: traverse the lithium-ion battery original data set, and determine the charging event and discharging event of each cycle according to the charge and discharge protocol, and whether the lithium-ion battery has experienced a complete charge or a complete discharge; S202: Based on the judgment result in step S201, the SOH value at the loop is calculated according to the following rules: If at cycle i, the lithium-ion battery has undergone a complete discharge, the calculation formula for SOH at cycle i is: If at cycle i, the lithium-ion battery has only undergone a complete charge and not a complete discharge, the calculation formula for SOH at cycle i is: If at cycle i, the lithium-ion battery has not undergone a complete charge and a complete discharge, then the SOH at cycle i is empty; Among them, t dc1 ,t dc2 The discharge start time and discharge end time are respectively, ch1 ,t ch2 They are the charging start time and the charging end time, I CH (t) is the charging current, I DC (t) is the discharge current, C rated is the rated capacity of the lithium-ion battery.

3. The semi-supervised SOH estimation method for lithium-ion batteries based on charging segments according to claim 1 is characterized in that: In step S3, data preprocessing is performed on the original data set of lithium-ion batteries to remove outliers in the data and fill missing values using interpolation. After data preprocessing, health feature extraction is performed.

4. The semi-supervised SOH estimation method for lithium-ion batteries based on charging segments according to claim 1 is characterized in that: Linearly interpolate the SOH between f1 and f2, and replace it with the mean and standard deviation of the interpolated curve and 5. The semi-supervised SOH estimation method for lithium-ion batteries based on charging segments according to claim 1 is characterized in that: The self-training recurrent neural network in step S5 includes: a self-training simple RNN neural network, a self-training LSTM neural network, and a self-training GRU neural network.

6. The method for semi-supervised SOH estimation of lithium-ion batteries based on charging segments according to claim 1, characterized in that: The self-training recurrent neural network adopts a self-training GRU neural network, which includes a GRU layer with 256 neurons, a GRU layer with 128 neurons, a fully connected layer with 32 neurons, and a fully connected layer with 1 neuron.

Images