Lithium ion battery health state estimation method, system and equipment based on image coding and FE-S-BiLSTM and medium

Abstract

The invention discloses a lithium ion battery health state estimation method, system and device based on image coding and FE-S-BiLSTM and a medium, and the method comprises the steps: collecting a current signal and a voltage signal in a battery charging process, and calculating a charging power signal; converting a charging power signal into a two-dimensional thermodynamic diagram by using an image coding technology, and extracting a battery health factor from the thermodynamic diagram in combination with structural similarity; a feature enhancement layer is introduced before an input layer of the BiLSTM, and an FS-BiLSTM model is generated; a reduction layer is introduced into the output part of the FS-BiLSTM to generate an FE-S-BiLSTM model; the FE-S-BiLSTM model is used as a lithium ion battery SOH estimation model; forming a data set by using the battery health factor and the battery SOH value, and training an FE-S-BiLSTM model by using the data set to obtain a trained FE-S-BiLSTM model; and estimating the health state of the lithium battery pack by using the trained FE-S-BiLSTM model. The aging essence of the lithium battery can be effectively reflected, and the health state of the lithium battery pack can be accurately estimated.

Description

Technical Field

[0001] The present invention relates to the field of battery technology, and in particular to a method, system, device and medium for estimating the health state of a lithium-ion battery based on image coding and FE-S-BiLSTM. Background Art

[0002] Lithium-ion batteries have become key energy storage components in power systems and have been widely used in consumer electronics due to their high energy density, high power density, low self-discharge rate, long service life and environmental protection. The aging factors of lithium-ion batteries are complex and diverse. As the battery charge and discharge cycles increase, the positive and negative electrode materials may undergo structural changes, and the loss of active lithium will also affect the battery capacity. In addition, the charging and discharging of the battery at different temperatures will affect its internal electrochemical reactions, while factors such as current rate, storage time, and charge and discharge depth will also affect battery performance. Although lithium-ion batteries dominate the field of new energy vehicles, their performance may increase the risk of accidents when it deviates from the ideal state. Accurately assessing the health status of lithium-ion batteries is crucial to ensuring the safety of electric vehicles. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a method, system, device and medium for estimating the health status of lithium-ion batteries based on image coding and FE-S-BiLSTM, which can effectively reflect the nature of lithium battery aging and accurately estimate the health status of lithium battery packs.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A method for estimating the health status of a lithium-ion battery based on image coding and FE-S-BiLSTM comprises the following steps:

[0006] S1, collecting the current signal and voltage signal during the battery charging process, and calculating the charging power signal;

[0007] S2. Use image coding technology to convert the charging power signal into a two-dimensional heat map, and extract the battery health factor from the heat map in combination with structural similarity;

[0008] S3, introduce a feature enhancement layer before the input layer of BiLSTM to generate the FS-BiLSTM model;

[0009] S4, introduce a reduction layer into the output part of FS-BiLSTM to generate FE-S-BiLSTM model;

[0010] S5. Use the FE-S-BiLSTM model as a lithium-ion battery SOH estimation model; form a data set with the battery health factor and the battery SOH value, and use the data set to train the FE-S-BiLSTM model to obtain a trained FE-S-BiLSTM model;

[0011] S6. Use the trained FE-S-BiLSTM model to estimate the health status of the lithium battery pack.

[0012] To optimize the above technical solutions, the specific measures taken also include:

[0013] Furthermore, S1 specifically includes:

[0014] Collecting the current signal and voltage signal of the whole charging scenario, and performing multiplication operation on the current signal and the voltage signal to obtain the charging power signal of the whole charging scenario;

[0015] The current signal and the voltage signal of the random segment charging scenario are collected, and the charging power signal of the random segment charging scenario is obtained by multiplying the current signal and the voltage signal.

[0016] Furthermore, in S2, the use of image coding technology to convert the charging power signal into a two-dimensional heat map is specifically as follows:

[0017] Using image coding technology, the power signal is mapped according to the viridis mapping criterion, and the charging power signal of the full charging scenario is converted into a heat map of the full charging scenario [T1, T2, ..., T n ], T n The heat map of the full charging scenario corresponding to the nth charging cycle is converted into the heat map form of the random segment charging scenario [T'1, T'2, ..., T' n ]; T' n A heat map showing a random segment charging scenario corresponding to the nth charging cycle;

[0018] The method of extracting the battery health factor from the heat map by combining structural similarity is as follows:

[0019] For the heat map of the full charging scenario, the heat map T1 of the full charging scenario corresponding to the first charging cycle is selected as the reference image T o , using structural similarity technology to quantify the thermal images of the full-stage charging scenarios with different cycle numbers and the baseline image T o The similarity between [β1,β2,…β i ,…,β n ], the formula is as follows:

[0020]

[0021] In the formula, β i is the similarity between the thermal image of the full charging scenario of the i-th cycle and the benchmark image, T x1 , and are the mean and variance of the window of the reference image and the window of the reference image, respectively, T xi , and are the window of the heat map of the full charging scenario of the i-th cycle number, the mean and variance of each window of the heat map; is the variance between the thermal image of the full-stage charging scenario at the i-th cycle number and the reference image; C1 is the first stable constant, and C2 is the second stable constant;

[0022] The similarity between the heat map of the full charging scenario and the benchmark image [β1,β2,…β i ,…,β n ] as the battery health factor obtained for the full charging scenario;

[0023] For the heat map of random segment charging scenarios, the heat map T'1 of the random segment charging scenario corresponding to the first charging cycle is selected as the reference image T' o , using structural similarity techniques to quantify the heatmaps of random fragment charging scenarios at different cycle times and the baseline image T' o The similarity between [β'1,β'2,…,β' i ,…,β' n ], the formula is as follows:

[0024]

[0025] Where β' i The heat map of the random charging scenario of the i-th charging cycle and the benchmark image T' o The similarity between x1 , and are the mean and variance of the window of the benchmark image and the window of the benchmark image for the random segment charging scenario, T' xi , and are the windows of the heat map of the random segment charging scenario, the mean and variance of each window, Variance between heatmaps for random segment charging scenarios and baseline images;

[0026] Compare the heat map of the random segment charging scenario with the benchmark image T' o The similarity between [β'1,β'2,…,β' i ,…,β' n] as the battery health factor obtained for random fragment charging scenarios.

[0027] Further, in S3, the feature enhancement layer is used to extract noise from the input battery health factor, and the feature enhancement layer includes a feature layer, an enhancement layer and an output layer;

[0028] The feature layer is used to map the input battery health factor, and the formula is as follows:

[0029]

[0030] Where Z i is the i-th battery health factor sample c i The feature mapping result, c i In the full-stage charging scenario, the battery health factor β obtained in the full-stage charging scenario is shown i , represents the battery health factor β' obtained in the random segment charging scenario under the random segment charging scenario i , where W ei , b ei are the weight and bias of the random mapping to the feature layer corresponding to the i-th battery health factor sample, and f(.) is the activation function of the feature layer;

[0031] The enhancement layer is used to increase the dimension of the input feature, and the formula is as follows:

[0032]

[0033] Among them, Z i is the i-th battery health factor sample c i The feature mapping result is is the feature mapping result Z i The corresponding connection weight when performing feature enhancement, is the feature mapping result Z i The corresponding bias when performing feature enhancement, H mi is the output of the enhancement layer; x(·) represents the activation function of the enhancement layer;

[0034] The output layer is used for feature fusion and outputs fusion features. The formula is as follows:

[0035] Y i =[Z i |H mi ]W i

[0036] Among them, Y i is the final output result of the output layer corresponding to the i-th battery health factor sample, and is the output result of the feature enhancement layer. iis the connection weight matrix when fusing the i-th battery health factor sample;

[0037] The BiLSTM part of the FS-BiLSTM model is used to capture the forward correlation feature K of the output result of the feature enhancement layer in the time space 1i and reverse correlation feature K 2i Finally, the forward and reverse correlation features are integrated to obtain the output result K of the FS-BiLSTM model i .

[0038] Furthermore, in S4, the reduction layer is used to remove redundant information and further select key features, and the formula is expressed as follows:

[0039]

[0040] In the formula, z ti is the output of the restoration gate, s represents the first activation function of the reduction layer, and r i is the output of the hidden state, L i-1 is the output value of the reduction layer at the previous moment, K i is the input of the reduction layer at the current moment, which comes from the output of the FS-BiLSTM model; b zi is the first bias of the reduction layer, b ri is the second bias of the reduction layer, b yi is the third bias of the reduction layer, represents the u-th row of the first weight matrix, represents the u-th row of the second weight matrix, W y represents the third weight matrix, tanh(·) represents the second activation function of the reduction layer; L i is the output value of the reduction layer at the current moment; is the intermediate state of the output value of the reduction layer.

[0041] Further, in S5, the specific process of forming a data set with the battery health factor and the battery SOH value and using the data set to train the FE-S-BiLSTM model is as follows:

[0042] The battery health factor [β1,β2,…β i ,…,β n ] and battery SOH values ​​[H1,H2,…H i , …, H n ], the training set and the test set are divided into 1:1 ratio in the time series space, where β i is the battery health factor of the full-stage charging scenario at the i-th cycle number, H iis the battery SOH value of the full-stage charging scenario at the i-th cycle number; the training set data is forward propagated into the FE-S-BiLSTM model. For each input training sample, the FE-S-BiLSTM model corresponds to an output result. The error between the model output result and the actual SOH value is calculated based on the loss function. The error is back-propagated in the FE-S-BiLSTM model, and the parameters in the FE-S-BiLSTM model are optimized based on the Adam optimization algorithm;

[0043] Repeat forward propagation, loss calculation, back propagation and parameter update to obtain a preliminarily trained FE-S-BiLSTM model;

[0044] The battery health factors obtained from the test set and random charging scenarios are used to verify the initially trained FE-S-BiLSTM model and fine-tune the model parameters to obtain a trained FE-S-BiLSTM model.

[0045] The present invention also proposes a lithium-ion battery health status estimation system based on image coding and FE-S-BiLSTM, comprising:

[0046] A data acquisition module is used to collect current and voltage signals during battery charging and calculate charging power signals;

[0047] The battery health factor extraction module is used to convert the charging power signal into a two-dimensional heat map using image coding technology, and extract the battery health factor from the heat map in combination with structural similarity;

[0048] The network modeling module is used to introduce a feature enhancement layer before the input layer of BiLSTM to generate a FS-BiLSTM model; introduce a reduction layer in the output part of FS-BiLSTM to generate a FE-S-BiLSTM model; and use the FE-S-BiLSTM model as a lithium-ion battery SOH estimation model;

[0049] The model training module is used to form a data set of battery health factors and battery SOH values, and use the data set to train the FE-S-BiLSTM model to obtain a trained FE-S-BiLSTM model;

[0050] The health status estimation module is used to estimate the health status of the lithium battery pack using the trained FE-S-BiLSTM model.

[0051] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the lithium-ion battery health status estimation method based on image coding and FE-S-BiLSTM as described above is implemented.

[0052] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the lithium-ion battery health state estimation method based on image coding and FE-S-BiLSTM as described above.

[0053] The beneficial effects of the present invention are:

[0054] (1) The battery health factor extraction scheme proposed in the present invention is based on the power in the charging stage as the test data. There are significant differences under different cycle numbers. The different numerical distributions are converted into color changes through image coding technology. The color changes of the heat map can more intuitively display the numerical size and change trend of the power during the charging process, which helps to capture the potential features and anomalies in the power signal caused by battery aging, and quantify such potential features and anomalies in combination with structural similarity, thereby enhancing the feature extraction capability.

[0055] (2) The battery SOH estimation model proposed in the present invention establishes a FE-S-BiLSTM model by introducing a feature enhancement layer and a simplification layer based on BiLSTM, wherein the feature enhancement layer mainly performs high-dimensional mapping on the extracted health factors by means of width expansion and incremental learning, thereby enriching the features and capturing the global static features. Then, based on the BiLSTM layer, the processed high-dimensional features are simultaneously captured in the past and future information, thereby improving the ability to capture long-term dependent information relationships in time series. Finally, the extracted complex features are simplified and integrated through the simplification layer, the key features are more prominent, and the global dynamic features are further efficiently captured.

[0056] (3) The method proposed in the present invention is not only applicable to the scenarios of full-stage charging and random-stage charging, but also can realize the application of the model trained based on the full-stage charging data of a battery combined with the battery SOH estimation method proposed in the present invention to other batteries with different working conditions. There is no need to re-establish the prediction model, which enhances the real-time monitoring and dynamic adjustment capabilities of the battery management system, improves the generalization ability of the evaluation model, and has significant practical application advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic flow chart of a method for estimating the health status of a lithium-ion battery based on FE-S-BiLSTM and image coding technology provided by an embodiment of the present invention;

[0058] Figure 2 It is the battery SOH change curve;

[0059] Figure 3 It is the comparison result of different estimation models based on the full-segment charging data;

[0060] Figure 4 It is the cross-battery SOH estimation result based on random fragment charging data;

[0061] Figure 5 It is the battery SOH estimation result across different battery charging scenarios. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0063] Embodiment 1

[0064] The present invention proposes a lithium-ion battery health state estimation method based on image coding and FE-S-BiLSTM. The overall process is as follows: Figure 1 As shown, the following steps are included:

[0065] S1, collecting the current signal and voltage signal during the battery charging process, and calculating the charging power signal;

[0066] In the embodiment of the present invention, in step S1, in the data collection work, the charging information of the x%-y% SOC interval is collected, which can be mainly divided into two charging scenarios. One is that x=0 and y=100, that is, the current signal of the full-segment charging scenario [Ii1, Ii2, ..., Ii j ] and charging voltage [Ui1,Ui2,...,Ui j ], and secondly, x and y are all values ​​between 0 and 100 and y is greater than x, that is, the current signal of the random segment charging scenario [I' i1 ,I' i2 ,...,I' ij ] and voltage signal [U' i1 ,U' i2 ,...,U' ij ], by multiplying the current signal and then the voltage signal, the charging power signal [P i1 ,P i2 ,...,P ij ] and [P' i1 ,P' i2 ,...,P' ij ], where I ij and U ij are the jth charging current and charging voltage samples in the i-th full-stage charging scenario, I' ij and U' ijare the jth charging current and charging voltage samples under the i-th random segment charging scenario, P ij and P' ij The jth charging power sample during the i-th charging process under the full-segment charging scenario and the random-segment charging scenario, n is the number of working cycles of the battery, and m is the number of samples collected under different charging scenarios, which is not a fixed value.

[0067] S2. Use image coding technology to convert the charging power signal into a two-dimensional heat map, and extract the battery health factor from the heat map in combination with structural similarity. In data collection, it is inevitable that the data contains noise. Therefore, the present invention converts the charging power time series data into a heat map. This processing method helps to capture potential features and anomalies in the power time series data.

[0068] The method of converting the charging power signal into a two-dimensional heat map using image coding technology is specifically as follows:

[0069] Using image coding technology, the power signal is mapped according to the viridis mapping criterion, and the charging power signal of the full charging scenario is converted into a heat map of the full charging scenario [T1, T2, ..., T n ], T n The heat map of the full charging scenario corresponding to the nth charging cycle is converted into the heat map form of the random segment charging scenario [T'1, T'2, ..., T' n ]; T' n A heat map showing a random segment charging scenario corresponding to the nth charging cycle;

[0070] The method of extracting the battery health factor from the heat map by combining structural similarity is as follows:

[0071] For the heat map of the full-stage charging scenario, the heat map T1 of the full-stage charging scenario corresponding to the first charging cycle is selected as the reference image T o , using structural similarity technology to quantify the thermal images of the full-stage charging scenarios with different cycle numbers and the baseline image T o The similarity between [β1,β2,…β i ,…,β n ], the formula is as follows:

[0072]

[0073] In the formula, β i is the similarity between the thermal image of the full charging scenario of the i-th cycle and the benchmark image, T x1 , and are the mean and variance of the window of the reference image and the window of the reference image, respectively, T xi , and are the window of the heat map of the full charging scenario of the i-th cycle number, the mean and variance of each window of the heat map; is the variance between the thermal image of the full-stage charging scenario at the i-th cycle number and the reference image; C1 is the first stable constant, and C2 is the second stable constant;

[0074] The similarity between the heat map of the full charging scenario and the benchmark image [β1,β2,…β i ,…,β n ] as the battery health factor obtained in the full charging scenario; thus quantifying the degree of battery aging. This method is relatively intuitive and easy to understand, and the health factor extracted by this scheme can very accurately characterize the battery SOH (battery health state).

[0075] For the heat map of random segment charging scenarios, the heat map T'1 of the random segment charging scenario corresponding to the first charging cycle is selected as the reference image T' o , using structural similarity techniques to quantify the heatmaps of random fragment charging scenarios at different cycle times and the baseline image T' o The similarity between [β'1,β'2,…,β' i ,…,β' n ], the formula is as follows:

[0076]

[0077] Where β' i The heat map of the random charging scenario of the i-th charging cycle and the benchmark image T' o The similarity between x1 , and are the mean and variance of the window of the benchmark image and the window of the benchmark image for the random segment charging scenario, T' xi , and are the windows of the heat map of the random segment charging scenario, the mean and variance of each window, Variance between heatmaps for random segment charging scenarios and baseline images;

[0078] Compare the heat map of the random segment charging scenario with the benchmark image T' o The similarity between [β'1,β'2,…,β' i ,…,β' n ] as the battery health factor obtained for random fragment charging scenarios.

[0079] S3. A feature enhancement layer is introduced before the input layer of BiLSTM to generate a FS-BiLSTM model; the feature enhancement layer is used to extract noise from the input battery health factor, and the feature enhancement layer includes a feature layer, an enhancement layer, and an output layer;

[0080] The feature layer is used to map the input battery health factor, and the formula is as follows:

[0081]

[0082] Where Z i is the i-th battery health factor sample c i The feature mapping result, c i In the full-stage charging scenario, the battery health factor β obtained in the full-stage charging scenario is shown i , represents the battery health factor β' obtained in the random segment charging scenario under the random segment charging scenario i , where W ei , b ei are the weight and bias of the random mapping to the feature layer corresponding to the i-th battery health factor sample, and f(.) is the activation function of the feature layer;

[0083] The enhancement layer is used to increase the dimension of the input feature, and the formula is as follows:

[0084]

[0085] Among them, Z i is the i-th battery health factor sample c i The feature mapping result is is the feature mapping result Z i The corresponding connection weight when performing feature enhancement, is the feature mapping result Z i The corresponding bias when performing feature enhancement, H mi is the output of the enhancement layer; x(·) represents the activation function of the enhancement layer;

[0086] The output layer is used for feature fusion and outputs fusion features. The formula is as follows:

[0087] Y i =[Z i |H mi ]W i

[0088] Among them, Y i is the final output result of the output layer corresponding to the i-th battery health factor sample, and is the output result of the feature enhancement layer. i is the connection weight matrix when fusing the i-th battery health factor sample;

[0089] The BiLSTM part of the FS-BiLSTM model is used to capture the forward correlation feature K of the output result of the feature enhancement layer in the time space 1i and reverse correlation feature K 2i Finally, the forward and reverse correlation features are integrated to obtain the output result K of the FS-BiLSTM model i .

[0090] S4. Introduce a reduction layer into the output part of FS-BiLSTM to generate a FE-S-BiLSTM model. The reduction layer is used to remove redundant information and further select key features. The formula is as follows:

[0091]

[0092] In the formula, z ti is the output of the restoration gate, s represents the first activation function of the reduction layer, and r i is the output of the hidden state, L i-1 is the output value of the reduction layer at the previous moment, K i is the input of the reduction layer at the current moment, which comes from the output of the FS-BiLSTM model; b zi is the first bias of the reduction layer, b ri is the second bias of the reduction layer, b yi is the third bias of the reduction layer, represents the u-th row of the first weight matrix, represents the u-th row of the second weight matrix, W y represents the third weight matrix, tanh(·) represents the second activation function of the reduction layer; L i is the output value of the reduction layer at the current moment; is the intermediate state of the output value of the reduction layer.

[0093] The FE-S-BiLSTM model simultaneously considers the impact of global static features and global dynamic features on the final SOH estimation results. The feature enhancement layer mainly uses width expansion and incremental learning methods to solve data quality problems caused by external uncertainties. Then, based on the simplification layer, it enhances its feature capture capability in the time series space and accurately captures the time series space features.

[0094] S5. Use the FE-S-BiLSTM model as a lithium-ion battery SOH estimation model. Use the battery health factor and the battery SOH value to form a data set, and use the data set to train the FE-S-BiLSTM model to obtain a trained FE-S-BiLSTM model. The specific process is as follows:

[0095] The battery health factor [β1,β2,…β i ,…,β n ] and battery SOH values ​​[H1,H2,…H i , …, H n ], the training set and the test set are divided into 1:1 ratio in the time series space, where β i is the battery health factor of the full-stage charging scenario at the i-th cycle number, H i is the battery SOH value of the full charging scenario at the i-th cycle number;

[0096] The training set contains [β1,β2,...,β n / 2 ] and [H1,H2,...,H n / 2 ], the test set contains [β (n / 2)+1 ,β (n / 2)+2 ,...,β n ] and [H (n / 2)+1 ,H (n / 2)+2 ,...,H n ];

[0097] The training set data is forward propagated into the FE-S-BiLSTM model. For each input training sample, the FE-S-BiLSTM model corresponds to an output result. The final model output test set corresponding battery SOH prediction result is [O1, O2, ..., O n / 2 ]; where O n / 2 Indicates the SOH prediction result of the n / 2th battery corresponding to the test set.

[0098] The error between the model output result and the actual SOH value is calculated based on the loss function. e1 ,O e2 ,...,O en / 2 ], the error is back-propagated in the FE-S-BiLSTM model, and the parameters in the FE-S-BiLSTM model are optimized based on the Adam optimization algorithm;

[0099] Repeat forward propagation, loss calculation, back propagation and parameter update to obtain a preliminarily trained FE-S-BiLSTM model;

[0100] The battery health factors obtained from the test set and random charging scenarios are used to verify the initially trained FE-S-BiLSTM model and fine-tune the model parameters to obtain a trained FE-S-BiLSTM model.

[0101] S6. Use the trained FE-S-BiLSTM model to estimate the health status of the lithium battery pack.

[0102] Embodiment 2

[0103] The present invention proposes a lithium-ion battery health status estimation system based on image coding and FE-S-BiLSTM corresponding to the method of embodiment 1, comprising:

[0104] A data acquisition module is used to collect current and voltage signals during battery charging and calculate charging power signals;

[0105] The battery health factor extraction module is used to convert the charging power signal into a two-dimensional heat map using image coding technology, and extract the battery health factor from the heat map in combination with structural similarity;

[0106] The network modeling module is used to introduce a feature enhancement layer before the input layer of BiLSTM to generate a FS-BiLSTM model; introduce a reduction layer in the output part of FS-BiLSTM to generate a FE-S-BiLSTM model; and use the FE-S-BiLSTM model as a lithium-ion battery SOH estimation model;

[0107] The model training module is used to form a data set of battery health factors and battery SOH values, and use the data set to train the FE-S-BiLSTM model to obtain a trained FE-S-BiLSTM model;

[0108] The health status estimation module is used to estimate the health status of the lithium battery pack using the trained FE-S-BiLSTM model.

[0109] The implementation methods of each module and module function in the system are completely consistent with the steps of the method in Example 1, so they will not be repeated here.

[0110] Embodiment 3

[0111] The present invention proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for estimating the health status of a lithium-ion battery based on image coding and FE-S-BiLSTM as described in Example 1 is implemented.

[0112] Embodiment 4

[0113] The present invention provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the lithium-ion battery health state estimation method based on image coding and FE-S-BiLSTM as described in the first embodiment.

[0114] In order to demonstrate the process and estimation performance of the lithium-ion battery health state estimation method based on image coding and FE-S-BiLSTM proposed in the present invention, an example is used here to illustrate.

[0115] The battery working data is collected based on the battery aging platform composed of a battery capacity tester (Neware BTS-400), a battery constant temperature environmental test chamber (Neware MHW-200) and a battery management system. Taking 18650 lithium batteries as the research object, the specific battery working conditions set this time are as follows: constant current charging with a current of 500mA until the voltage across the battery reaches a cut-off voltage of 4.2V, and then constant voltage charging with a constant voltage of 4.2V until the current drops to 48mA, stop charging, and discharge at a constant current of 1250mA until the voltage drops to 3V. This experiment simulated full-segment charging and three random segment charging scenarios, among which the three random segment charging scenarios were charging SOC segment ranges of 53%-81%, 63%-95% and 73%-100%. The battery SOH is estimated based on the collected charging data, and the specific process is as follows:

[0116] (1) The power information of the battery charging process is collected based on the laboratory battery aging platform, which contains the power signals of the full-segment charging information and random-segment charging information. The SOH of the lithium battery is as follows: Figure 2 shown.

[0117] (2) The charging power signals of different cycle times in the full-segment charging and random-segment charging scenarios are converted into heat maps using data encoding technology. In both the full-segment charging scenario and the random-segment charging scenario, the heat map corresponding to the charging information in the first cycle is selected as the reference image.

[0118] (3) Use structural similarity to quantify the similarity between the heat map corresponding to different charging scenarios and the reference image. The Pearson correlation coefficient method is used to quantify the correlation between the health factors extracted in different charging scenarios and the battery SOH. Through calculation, the correlation coefficient between the health factors proposed based on the full-segment charging data and the battery SOH is 0.99881, and the correlation coefficient between the health factors extracted based on random segment charging data and the battery SOH is greater than 0.99. The results can preliminarily prove that the battery health factor extraction scheme proposed in the present invention has strong practicality, whether it is based on the full-segment data or not, it can accurately reflect the aging nature of the battery.

[0119] (4) The extracted health factors are used as input features, and the battery SOH is used as the target variable. The training set and the test set are divided into a 1:1 ratio in the time series space, that is, the first half of the cycles are the training set, and the latter half of the cycles are the test set.

[0120] (5) FE-S-BiLSTM evaluation models are established for full charging scenarios and random charging scenarios, and the RMSE, MAE and R 2The model is comprehensively evaluated by indicators such as Figure 3 and Figure 4 In the full charging scenario, the RMSE, MAE and R between the final SOH estimate and the true value 2 They are 0.43133%, 0.33100% and 0.98347 respectively. The corresponding evaluation indicators of the prediction results based on random fragments are shown in Table 1.

[0121] Table 1 SOH estimation performance indicators for random segment charging data

[0122] SOC range <![CDATA[R 2 ]]> RMSE(%) MAE(%) 53％-81％ 0.93742 1.05606 0.91694 63％-95％ 0.90440 1.06431 1.01820 73％-100％ 0.94696 0.84505 0.70105

[0123] (6) Based on the trained battery SOH estimation model and combined with the full charging data and random segment charging data of another single battery, the battery SOH estimation task across multiple battery charging scenarios is realized. The specific working conditions of this single battery are different from those described previously. The final experimental results are shown in the figure below. Figure 5 The final model evaluation indicators are shown in Table 2.

[0124] Table 2 Cross-battery SOH evaluation indicators

[0125] SOC range R2 RMSE(%) MAE(%) 31％-83％ 0.970491 0.95481 0.86458 47％-87％ 0.940174 1.44924 1.17629 64％-100％ 0.95186 1.29121 1.00978

[0126] from Figure 2 It can be clearly observed that the accuracy of the model proposed in the present invention in the battery SOH estimation task is much higher than that of the other comparison models, and it has better tracking ability in the part where the battery SOH fluctuates more violently. It is further verified that the model captures the static features in the data through the feature enhancement layer, uses BiLSTM bidirectional information propagation to effectively capture the time series data features, and finally further extracts and integrates the global dynamic features through the simplification layer. This method significantly enhances the fitting and generalization capabilities of the model, and effectively improves the accuracy and generalization capabilities of the battery SOH estimation. At the same time, the model proposed in the present invention exhibits strong model estimation performance in both random charging scenarios and cross-battery SOH estimation tasks. The effectiveness and superiority of this method are further demonstrated through examples, providing a scientific basis for accurately evaluating the SOH status.

[0127] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0128] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0129] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A lithium-ion battery health state estimation method based on image coding and FE-S-BiLSTM, characterized in that: The following steps are involved: S1, collecting the current signal and voltage signal during the battery charging process, and calculating the charging power signal; S2. Use image coding technology to convert the charging power signal into a two-dimensional heat map, and extract the battery health factor from the heat map in combination with structural similarity; S3, introduce a feature enhancement layer before the input layer of BiLSTM to generate the FS-BiLSTM model; S4, introduce a reduction layer into the output part of FS-BiLSTM to generate FE-S-BiLSTM model; S5. Use the FE-S-BiLSTM model as a lithium-ion battery SOH estimation model; form a data set with the battery health factor and the battery SOH value, and use the data set to train the FE-S-BiLSTM model to obtain a trained FE-S-BiLSTM model; S6. Use the trained FE-S-BiLSTM model to estimate the health status of the lithium battery pack.

2. The method for estimating the health status of a lithium-ion battery based on image coding and FE-S-BiLSTM as claimed in claim 1, characterized in that: S1 specifically includes: Collecting the current signal and voltage signal of the whole charging scenario, and performing multiplication operation on the current signal and the voltage signal to obtain the charging power signal of the whole charging scenario; The current signal and the voltage signal of the random segment charging scenario are collected, and the charging power signal of the random segment charging scenario is obtained by multiplying the current signal and the voltage signal.

3. The method for estimating the health status of a lithium-ion battery based on image coding and FE-S-BiLSTM as claimed in claim 1, characterized in that: In S2, the use of image coding technology to convert the charging power signal into a two-dimensional heat map is specifically as follows: Using image coding technology, the power signal is mapped according to the viridis mapping criterion, and the charging power signal of the full charging scenario is converted into a heat map of the full charging scenario [T1, T2, ..., T n ], T n The heat map of the full charging scenario corresponding to the nth charging cycle is converted into the heat map form of the random segment charging scenario [T'1, T'2, ..., T' n ]; T' n A heat map showing a random segment charging scenario corresponding to the nth charging cycle; The method of extracting the battery health factor from the heat map by combining structural similarity is as follows: For the heat map of the full-stage charging scenario, the heat map T1 of the full-stage charging scenario corresponding to the first charging cycle is selected as the reference image T o , using structural similarity technology to quantify the thermal images of the full-stage charging scenarios with different cycle numbers and the baseline image T o The similarity between [β1,β2,…β i ,…,β n ], the formula is as follows: In the formula, β i is the similarity between the thermal image of the full charging scenario of the i-th cycle and the benchmark image, T x1 , and are the mean and variance of the window of the reference image and the window of the reference image, respectively, T xi , and are the window of the heat map of the full charging scenario of the i-th cycle number, the mean and variance of each window of the heat map; is the variance between the thermal image of the full-stage charging scenario at the i-th cycle number and the reference image; C1 is the first stable constant, and C2 is the second stable constant; The similarity between the heat map of the full charging scenario and the benchmark image [β1,β2,…β i ,…,β n ] as the battery health factor obtained for the full charging scenario; For the heat map of random segment charging scenarios, the heat map T'1 of the random segment charging scenario corresponding to the first charging cycle is selected as the reference image T' o , using structural similarity techniques to quantify the heatmaps of random fragment charging scenarios at different cycle times and the baseline image T' o The similarity between [β'1,β'2,…,β' i ,…,β' n ], the formula is as follows: Where β' i The heat map of the random charging scenario of the i-th charging cycle and the benchmark image T' o The similarity between x1 , and are the mean and variance of the window of the benchmark image and the window of the benchmark image for the random segment charging scenario, T' xi , and are the windows of the heat map of the random segment charging scenario, the mean and variance of each window, Variance between heatmaps for random segment charging scenarios and baseline images; Compare the heat map of the random segment charging scenario with the benchmark image T' o The similarity between [β'1,β'2,…,β' i ,…,β' n ] as the battery health factor obtained for random fragment charging scenarios.

4. The method for estimating the health status of a lithium-ion battery based on image coding and FE-S-BiLSTM as claimed in claim 1, characterized in that: In S3, the feature enhancement layer is used to extract noise from the input battery health factor, and the feature enhancement layer includes a feature layer, an enhancement layer and an output layer; The feature layer is used to map the input battery health factor, and the formula is as follows: Where Z i is the i-th battery health factor sample χ i The feature mapping result, χ i In the full-stage charging scenario, the battery health factor β obtained in the full-stage charging scenario is shown i , represents the battery health factor β' obtained in the random segment charging scenario under the random segment charging scenario i , where W ei , b ei are the weight and bias of the random mapping to the feature layer corresponding to the i-th battery health factor sample, and φ(.) is the activation function of the feature layer; The enhancement layer is used to increase the dimension of the input feature, and the formula is as follows: Among them, Z i is the i-th battery health factor sample χ i The feature mapping result is is the feature mapping result Z i The corresponding connection weight when performing feature enhancement, is the feature mapping result Z i The corresponding bias when performing feature enhancement, H mi is the output of the enhancement layer; ξ(·) represents the activation function of the enhancement layer; The output layer is used for feature fusion and outputs fusion features. The formula is as follows: Y i =[Z i |H mi ]W i Among them, Y i is the final output result of the output layer corresponding to the i-th battery health factor sample, as the output result of the feature enhancement layer, W i is the connection weight matrix when fusing the i-th battery health factor sample; The BiLSTM part of the FS-BiLSTM model is used to capture the forward correlation feature K of the output result of the feature enhancement layer in the time space 1i and reverse correlation feature K 2i Finally, the forward and reverse correlation features are integrated to obtain the output result K of the FS-BiLSTM model i .

5. The method for estimating the health status of a lithium-ion battery based on image coding and FE-S-BiLSTM as claimed in claim 1, characterized in that: In S4, the reduction layer is used to remove redundant information and further select key features. The formula is as follows: In the formula, z ti is the output of the restoration gate, σ represents the first activation function of the reduction layer, r i is the output of the hidden state, L i-1 is the output value of the reduction layer at the previous moment, K i is the input of the reduction layer at the current moment, which comes from the output of the FS-BiLSTM model; b zi is the first bias of the reduction layer, b ri is the second bias of the reduction layer, b yi is the third bias of the reduction layer, represents the u-th row of the first weight matrix, represents the u-th row of the second weight matrix, W y represents the third weight matrix, tanh(·) represents the second activation function of the reduction layer; L i is the output value of the reduction layer at the current moment; is the intermediate state of the output value of the reduction layer.

6. The method for estimating the health status of a lithium-ion battery based on image coding and FE-S-BiLSTM as claimed in claim 1, characterized in that: In S5, the specific process of forming a data set with the battery health factor and the battery SOH value and using the data set to train the FE-S-BiLSTM model is as follows: The battery health factor [β1,β2,…β i ,…,β n ] and battery SOH values ​​[H1,H2,…H i , …, H n ], the training set and the test set are divided into 1:1 ratio in the time series space, where β i is the battery health factor of the full-stage charging scenario at the i-th cycle number, H i is the battery SOH value of the full-stage charging scenario of the i-th cycle; the training set data is forward propagated into the FE-S-BiLSTM model. For each input training sample, the FE-S-BiLSTM model corresponds to an output result. The error between the model output result and the actual SOH value is calculated based on the loss function. The error is back-propagated in the FE-S-BiLSTM model, and the parameters in the FE-S-BiLSTM model are optimized based on the Adam optimization algorithm; Repeat forward propagation, loss calculation, back propagation and parameter update to obtain a preliminarily trained FE-S-BiLSTM model; The battery health factors obtained from the test set and random charging scenarios are used to verify the initially trained FE-S-BiLSTM model and fine-tune the model parameters to obtain a trained FE-S-BiLSTM model.

7. A lithium-ion battery health status estimation system based on image coding and FE-S-BiLSTM, characterized in that: include: A data acquisition module is used to collect current and voltage signals during battery charging and calculate charging power signals; The battery health factor extraction module is used to convert the charging power signal into a two-dimensional heat map using image coding technology, and extract the battery health factor from the heat map in combination with structural similarity; The network modeling module is used to introduce a feature enhancement layer before the input layer of BiLSTM to generate a FS-BiLSTM model; Introduce a reduction layer into the output part of FS-BiLSTM to generate a FE-S-BiLSTM model; use the FE-S-BiLSTM model as a lithium-ion battery SOH estimation model; The model training module is used to form a data set of battery health factors and battery SOH values, and use the data set to train the FE-S-BiLSTM model to obtain a trained FE-S-BiLSTM model; The health state estimation module is used to estimate the health state of the lithium battery pack using the trained FE-S-BiLSTM model.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for estimating the health status of a lithium-ion battery based on image coding and FE-S-BiLSTM as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables a computer to execute the lithium-ion battery health state estimation method based on image coding and FE-S-BiLSTM as described in any one of claims 1 to 6.

Images