A lithium ion battery state of charge and state of health combined estimation method and system

Abstract

The present application belongs to the technical field of battery state of charge estimation, and provides a lithium ion battery state of charge and health state combined estimation method and system, obtains the capacity of the battery in the charging and discharging cycle process and the stress and temperature data at different positions; the obtained data is pretreated; the spirlman rank correlation coefficient of each stress, temperature point, state of charge and health state after pretreatment is calculated and analyzed respectively, the stress and temperature characteristics which change more than a set degree and are associated with the battery state more than a set value are screened out as key characteristics; the screened key characteristics are taken as input, the state of charge value and the health state value are taken as output, the model loss function is set, and the combined estimation model is trained; the key characteristics of the real-time obtained stress and temperature data are input into the trained combined estimation model, and the state of charge and health state estimation values are obtained. The present application realizes high-precision estimation of SOC and SOH.

Description

Technical Field

[0001] This invention belongs to the field of battery state of charge estimation technology, specifically relating to a method and system for jointly estimating the state of charge and health of lithium-ion batteries. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] State of Charge (SOC) is a key parameter characterizing the internal properties and operating state of a battery, directly affecting its performance, safety, and lifespan. State of Health (SOH) reflects the degree of performance degradation, directly impacting its safety, reliability, and lifespan. The two are interdependent and closely related. Therefore, accurate SOC and SOH estimation is a core task in battery management systems, crucial for achieving efficient, stable, and safe battery operation.

[0004] To achieve accurate joint estimation of battery SOC and SOH, scholars both domestically and internationally have conducted extensive research. This mainly includes model-based methods and data-driven approaches. Some studies have established a fractional-order equivalent circuit model of the battery, used pulse charge-discharge data and a genetic algorithm to determine model parameters, and combined this with a double extended Kalman filter algorithm to achieve joint estimation of SOC and SOH. However, this method is highly complex, computationally intensive, and heavily dependent on the model, making it susceptible to external influences and resulting in significant estimation errors.

[0005] Other literature has built a joint battery state estimation model based on the TCN-LSTM model and optimized the model hyperparameters using a Bayesian optimization algorithm, enabling the model to simultaneously estimate SOC and SOH. However, this method only considers electrochemical data such as voltage and current, making it susceptible to measurement noise and external environmental influences. This results in the model being difficult to adapt to various operating conditions, and its robustness and accuracy are difficult to further improve. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a method and system for jointly estimating the state of charge (SOC) and state of health (SOH) of lithium-ion batteries. This invention collects multiple stress and temperature data on the battery surface during use in a distributed, real-time manner, analyzes and extracts the stress difference data between the positive and negative electrode regions of the battery as feature data, and builds a joint SOC-SOH estimation model for the battery by combining two Transformer models with different time scales, thereby achieving high-precision joint estimation of multiple battery states.

[0007] According to some embodiments, the present invention adopts the following technical solution:

[0008] A method for jointly estimating the state of charge and state of health of a lithium-ion battery includes the following steps:

[0009] Obtain the capacity of the lithium-ion battery during the charge-discharge cycle, as well as the stress and temperature data at different locations, recorded in the lithium-ion battery charge-discharge experiment.

[0010] The acquired data is preprocessed;

[0011] Spearman's rank correlation coefficients between each stress and temperature point after preprocessing and the state of charge and health were calculated and analyzed. Stress and temperature features that changed beyond a set level and were correlated with the battery state beyond a set value were selected as key features.

[0012] The selected key features are used as inputs, and the state of charge and health values ​​are used as outputs. The model loss function is set, and the joint estimation model is trained.

[0013] By inputting key features of real-time acquired stress and temperature data into the trained joint estimation model, the final estimates of the state of charge and health status are obtained.

[0014] As an alternative implementation, the stress data includes a positive electrode tab region, a negative electrode tab region, and an intermediate region located between the two tab regions, and the intermediate region is obtained by dividing it into multiple regional partitions.

[0015] As an alternative implementation, the temperature data includes a positive electrode tab region, a negative electrode tab region, and an intermediate region located between the two tab regions, and the intermediate region is divided into multiple regional partitions for acquisition.

[0016] As an alternative implementation method, the process of calculating and analyzing the Spearman-level correlation coefficients of each pre-processed stress and temperature point with the state of charge and the state of health, and selecting stress and temperature features that change beyond a set level and are correlated with the battery state beyond a set value as key features includes: calculating the correlation between stress difference and temperature in each region with the state of charge and the state of health; using the average temperature data of each location region within a set range from the negative electrode tab region, and the stress difference between the positive and negative electrode tabs as key features for estimating the state of charge; and using the maximum and minimum values ​​of the stress difference between the positive and negative electrode tabs as key features for estimating the state of health.

[0017] As an alternative implementation, the joint estimation model includes two Transformer sub-models. For the Transformer sub-model used for state of charge estimation, the coupling relationship with the health state is preserved. The input features are the stress difference Δσ between the positive and negative electrode tabs, the average temperature data of the negative electrode tab region, and the health state data. The output is the estimated state of charge value.

[0018] For the Transformer sub-model used for health status estimation, the input features are the maximum and minimum values ​​of the stress difference Δσ between the positive and negative electrodes in each cycle, and the output features are the health status estimates.

[0019] As an alternative implementation, each Transformer sub-model includes an input layer, a feature capture module, and an output layer. The input layer is used to acquire key feature data of the input and encode its position to combine sequence information with the input data.

[0020] The feature capture module is used to extract the global dependency relationship between each feature point in the sequence and other points, and to introduce nonlinear feature transformation;

[0021] The output layer is used to integrate the dimensions of the output of the feature capture module into the required target value and output the state of charge estimation result.

[0022] As an alternative implementation, the feature capture module includes a multi-layered stacked encoder. Each encoder includes a multi-head attention mechanism layer, a summation and normalization layer, a feedforward fully connected network, and a summation and normalization layer. The multi-head attention mechanism layer is used to extract the global dependencies between each feature point and other points in the sequence, enabling each position to interact with all other positions and dynamically weighted according to the attention score. The summation and normalization layer is used to prevent gradient vanishing in deep networks, accelerate convergence, and improve model stability. The feedforward fully connected network includes two fully connected neural network layers, which are used to independently introduce nonlinear feature transformations at each position, enhancing the model's expressive power.

[0023] As an alternative implementation, the encoder has 4 layers, the activation function is ReLU, the optimizer is Adam, and the loss function is mean squared error.

[0024] As an alternative implementation method, the process of setting the model loss function includes: For SOC estimation, the loss function is designed as follows:

[0025] ;

[0026] The meanings of each variable are as follows:

[0027] The basic mean square error term, SOC pred Represents the SOC estimate. SOC true Represents the true value of SOC. αUsing the mean squared error as the weighting coefficient for this term, the instantaneous fluctuations of SOC can be accurately and sensitively captured.

[0028] β •ReLU(Δ σ - σ th ) represents the exponential stress risk penalty term, Δ σ This represents the stress difference between the positive and negative electrode tab regions. σ th ReLU represents the stress safety threshold, which is set according to the yield strength of the specific battery material. ReLU represents a linear threshold function. β This is the weighting coefficient for this item, adjusted as follows: , The reference value is set for use in Δ σ When the stress difference is too large, the weight is increased and the penalty is increased. The purpose of this design is to reflect that when the stress difference exceeds the safety threshold, the penalty is activated, forcing the model to pay attention to the high stress risk.

[0029] δ •∣▽ T | represents the uniform thermal gradient term, ▽ T This represents the temperature difference between the tab region and the central region, reflecting the degree of non-uniformity in the temperature field. δ This is the weighting coefficient for this item, calculated as follows: , T ref A temperature is set at the start of the cycle, and the weight gradually increases as the temperature rises. The baseline value is set, and the purpose of this design is to minimize temperature distribution differences and suppress the risk of local thermal runaway;

[0030] This is an electromechanical response constraint term, designed based on the observed strong correlation between battery stress and voltage variation trends. Represents the rate of change of voltage. Represents the rate of stress change, and k is the voltage-stress coupling coefficient. μ The coupling weight is calculated as follows: , The set reference value; stress gradient ▽ σ The larger the variance, μ It increases linearly, taking into account the synergy of multiple physical fields in the battery to ensure that voltage changes are synchronized with stress changes;

[0031] For SOH estimation, the loss function is designed as follows:

[0032] ;

[0033] in, Based on the aging error term, SOH pred Represents the estimated SOH value. SOH true Representing the true SOH value, this term serves as the basic error term for SOH estimation. It can adapt to the characteristics of slow battery aging, minimize the absolute error of SOH estimation, and avoid interference from outliers in a single cycle with the long-term degradation trend.

[0034] This is a stress-internal resistance correlation term, designed based on the strong correlation between stress difference and internal resistance in each cycle. This represents the ratio of the stress difference amplitude between the nth cycle and the first cycle. This represents the ratio of the internal resistance of the nth cycle to that of the 1st cycle; λ The association weight is calculated as follows: , The set baseline value is an aging consistency constraint that forces the increase in stress difference amplitude to be synchronized with the increase in internal resistance, and strengthens the correlation between stress and chemical aging in the later stages of battery aging.

[0035] This is a voltage relaxation-aging coupling term. V relax This represents the static voltage drop. Represents the SOH decay rate. η This is the coupling coefficient, calculated as follows: , When the static pressure drop deviates from the set normal value (based on the baseline value), η The voltage increases in a stepwise manner. This design aims to force the SOH to accelerate its decay when the voltage relaxation increases, which is consistent with the electrochemical aging characteristics of the battery.

[0036] A joint estimation system for the state of charge and state of health of a lithium-ion battery includes:

[0037] The data acquisition module is configured to acquire the capacity of the battery during the charge-discharge cycle, as well as the stress and temperature data at different locations, recorded in the lithium-ion battery charge-discharge experiment.

[0038] The preprocessing module is configured to preprocess the acquired data;

[0039] The correlation analysis module is configured to calculate and analyze the Spearman-level correlation coefficients between each pre-processed stress and temperature point and the state of charge and health, and to select stress and temperature features that change beyond a set degree and are correlated with the battery state beyond a set value as key features.

[0040] The model training module is configured to take the selected key features as input, the state of charge and health values ​​as output, set the model loss function, and train the joint estimation model.

[0041] The estimation module is configured to input key features of real-time acquired stress and temperature data into the trained joint estimation model to obtain the final estimates of the state of charge and health.

[0042] As an alternative implementation, the data acquisition module includes a fixing clamp, a thin-film sensor, a load cell, and a heat insulation pad. The thin-film sensor is laid on the heat insulation pad, and a lithium-ion battery is placed on the thin-film sensor. The fixing clamp clamps the three together, so that the thin-film sensor is in close contact with the surface of the lithium-ion battery. The load cell is used to collect the preload applied to the lithium-ion battery. The thin-film sensor detects the stress and temperature data at various locations on the lithium-ion battery.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] This invention not only fully considers the differences in stress and temperature characteristics in different regions of the battery, but also reduces computational complexity and redundant information through feature selection, while adapting to different types and structures of lithium-ion batteries. Furthermore, by utilizing the established estimation model, it further extracts the complex dependencies between multidimensional features, resulting in high computational efficiency, strong flexibility and interpretability, and significantly improving the accuracy and robustness of SOC estimation.

[0045] This invention uses thin-film sensors to collect multiple stress and temperature data in a distributed manner, which can more comprehensively reflect the thermal change characteristics of different locations on the battery. The measurement of temperature and stress characteristics on the battery surface is simple and convenient, making it easier to apply in practice.

[0046] This invention selects the key stress and temperature data that best characterize SOC as input features, enabling the model to more fully explore the complex nonlinear relationship between the data and SOC, and achieve a more stable and reliable SOC estimation.

[0047] This invention builds a joint estimation model based on Transformer encoder structures at different time scales, which can adapt to different time scales between battery SOC and SOH, improve computational efficiency, fully capture the dependencies between multidimensional features, and is highly flexible with accurate estimation results.

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0049] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0050] Figure 1 This is a flowchart illustrating the joint estimation of battery state of charge and state of health in one embodiment.

[0051] Figure 2 A model structure diagram of one embodiment;

[0052] Figure 3 This is a diagram of a joint estimation model for SOC-SOH in one embodiment;

[0053] Figure 4 This is a schematic diagram of a distributed stress-temperature integrated acquisition device according to one embodiment;

[0054] Figure 5 This is a schematic diagram illustrating stress changes and region division during battery charging in one embodiment.

[0055] Figure 6 This is a schematic diagram of the average temperature of each region in one embodiment;

[0056] Figure 7 This is a schematic diagram of the stress difference Δσ variation curve during a cycle in one embodiment.

[0057] Figure 8 This is a schematic diagram of SOC estimation results and errors in one embodiment, where (a) is a schematic diagram of estimation results and (b) is a schematic diagram of errors;

[0058] Figure 9 This is a schematic diagram of SOH estimation results and errors in one embodiment, where (a) is a schematic diagram of estimation results and (b) is a schematic diagram of errors;

[0059] Figure 10 This is a flowchart illustrating the operation of the loss function in one embodiment. Detailed Implementation

[0060] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0061] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0062] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0063] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0064] During the charging and discharging process of a battery, lithium ions continuously migrate between the positive electrode, negative electrode, and electrolyte, causing periodic changes in the battery's mechanical stress. During charging, lithium ions escape from the positive electrode, diffuse through the electrolyte, and embed themselves into the negative electrode, causing the negative electrode material to expand and thus increasing the overall battery stress. Conversely, during discharging, lithium ions escape from the negative electrode and re-embed into the positive electrode, causing the negative electrode material to shrink and the battery stress to decrease. This stress, which changes repeatedly with charge-discharge cycles, is called reversible stress, and its amplitude and trend effectively reflect the battery's state of charge.

[0065] With prolonged use and aging, irreversible physical and chemical processes such as the continuous growth of the solid electrolyte interphase (SEI) film, the decomposition of the electrolyte producing gas, and the formation of lithium dendrites lead to the accumulation of irreversible stress within the battery. This irreversible stress can cause mechanical damage, particle breakage, and electrode delamination, resulting in structural degradation and ultimately affecting battery safety and lifespan. Therefore, stress changes are not only an important parameter characterizing battery state evolution but also play a crucial role in battery state management and safety assessment. In battery state estimation and health management, fully considering stress characteristics is essential for improving the accuracy of SOC and SOH estimations and ensuring safe battery operation.

[0066] Furthermore, temperature characteristics also change during battery cycling. Temperature directly affects the rate of electrochemical reactions, internal resistance, and electrolyte activity within the battery, thus significantly altering its charge / discharge capacity and voltage characteristics. Low temperatures reduce lithium-ion migration speed and generate side reactions, leading to reduced usable capacity and inflated State of Charge (SOC); high temperatures may cause localized overheating, accelerate battery material aging, and trigger capacity decay and safety risks. Ignoring temperature compensation can increase SOC estimation errors, reduce the accuracy of the battery management system, and even lead to safety hazards such as overcharging / over-discharging. Therefore, it is also necessary to consider battery temperature characteristics during SOC estimation.

[0067] Based on this, the present invention proposes a joint estimation method for the state of charge and state of health of lithium-ion batteries, particularly a joint estimation method for the state of charge and state of health of lithium-ion batteries based on distributed stress and temperature characteristics, such as... Figure 1 As shown, it includes the following steps:

[0068] (1) Conduct charge and discharge experiments on the battery and record the battery capacity and stress and temperature data at different locations during the charge and discharge cycle.

[0069] (2) Preprocess the data to improve data quality and reduce the impact of environmental noise;

[0070] (3) Calculate and analyze the stress difference Δσ in the positive and negative electrode tab regions respectively. - -Δσ + Spearman correlation of negative electrode tab temperature T with SOC and SOH, stress and temperature features that are strongly correlated with battery state are selected as input features.

[0071] (4) Build Transformer sub-models at different time scales for SOC and SOH respectively, jointly construct a joint estimation model, construct and set the model loss function, and then use the selected features as input and SOC and SOH values ​​as output to train the model;

[0072] (5) Input the stress and temperature features collected in real time into the trained model and output the estimated values ​​of SOC and SOH to achieve accurate estimation of battery state.

[0073] Traditional Transformers consist of an encoder and a decoder. For long sequence processing tasks with multiple feature inputs and a single feature output, feature modeling and prediction can be achieved using only the encoder. Therefore, to better utilize extracted stress and temperature features to estimate battery state, this invention builds a base model based on the encoder module of a Ttransformer, as follows: Figure 2 As shown, the main structure and function of the model are as follows:

[0074] (1) Input part: including input feature data and positional encoding. The input features are the average stress values ​​and temperature data of regions B, O and P. Positional encoding can combine sequence information with input data, enabling the model to learn time series information. Here, sine and cosine encoding is used.

[0075] (2) Feature Capture Part: This invention employs a 4-layer stacked encoder. Among them, the multi-head self-attention mechanism in the encoder is used to extract the global dependency relationship between each feature point in the sequence and other points, so that each position can interact with all other positions and is dynamically weighted according to the attention score; the summation and normalization (Add&Nom) module is used to prevent gradient vanishing in deep networks, accelerate convergence, and improve model stability; the feed forward network (FFN) structure is a two-layer fully connected neural network, which allows each position to independently introduce nonlinear feature transformations, enhancing the model's expressive power.

[0076] (3) Output section: includes fully connected layer and output layer. The fully connected layer can integrate the output of the high-dimensional encoder into the required target value; the output layer is a linear regression layer, which outputs the SOC estimation result.

[0077] During battery use, the State of Charge (SOC) changes in real time, while the State of Harm (SOH) is typically measured in cycles. To accommodate the different time scales between SOC and SOH, this invention combines two Transformer sub-models to build a joint SOC-SOH estimation model, such as... Figure 3 As shown. Furthermore, for the SOC estimation module, the coupling relationship with SOH is preserved. The input features are the real-time acquired stress difference Δσ, the average temperature data of the negative electrode tab region C, and the health status data SOH. The output feature is the estimated SOC value, i.e., SOC. t+1 =[Δσ t , T t SOH t The sampling frequency is t seconds. For the SOH estimation module, the input features are the maximum and minimum values ​​of the stress difference Δσ in each cycle, and the output feature is the estimated SOH value, i.e., SOH. n+1 =[ , The sampling frequency is the number of iterations n. In this invention, each Transformer model has 4 encoder layers, ReLU is used as the activation function, and Adam is used as the optimizer.

[0078] In addition, such as Figure 10As shown, this invention introduces a physically constrained hierarchical loss function architecture into the multi-state estimation of the Transformer architecture. The designed loss function includes the loss function of the SOC estimation module and the loss function of the SOH estimation module. This method does not require changing the main structure of the neural network, constructs multi-physics constraints through easily obtainable indicators such as stress, temperature, and voltage, and combines a dynamic weight adjustment mechanism to improve the reliability and generalization ability of the prediction, demonstrating good universality and innovation. The composition of each loss function and the meaning of its variables are explained in detail below.

[0079] For the SOC estimation module (second-level), the loss function is designed as follows:

[0080] ;

[0081] The meanings of each variable are as follows:

[0082] The basic mean square error term, SOC pred Represents the SOC estimate. SOC true Represents the true value of SOC. α This is the weighting coefficient for this item, with a default value of 1. Choosing the mean squared error as the basic error term for SOC estimation can accurately and sensitively capture the instantaneous fluctuations of SOC.

[0083] β •ReLU(Δ σ - σ th ) represents the exponential stress risk penalty term, Δ σ This represents the stress difference between the positive and negative electrode tab regions. σ th ReLU represents the stress safety threshold, which is set according to the yield strength of the specific battery material. ReLU represents a linear threshold function. β This is the weighting coefficient for this item, with a default value of 0.5. The adjustment method is as follows: , The reference value is set for use in Δ σ When the stress difference is too large, the weight is increased to increase the penalty. This design reflects the activation of the penalty when the stress difference exceeds the safety threshold, forcing the model to focus on high stress risks.

[0084] δ •∣▽ T | represents the uniform thermal gradient term, ▽ T This represents the temperature difference between the tab region and the central region, reflecting the degree of non-uniformity in the temperature field. δ This is the weighting coefficient for this item, with a default value of 0.3. The calculation method is as follows: , T refA temperature is set at the start of the cycle, and the weight gradually increases as the temperature rises. This is the baseline value set. The purpose of this design is to minimize temperature distribution differences and suppress the risk of localized thermal runaway.

[0085] This is an electromechanical response constraint term, designed based on the observed strong correlation between battery stress and voltage variation trends. Represents the rate of change of voltage (V / s). The stress change rate (MPa / s) is represented by k, which is the voltage-stress coupling coefficient. μ This is the coupling weight for this item, with a default value of 0.1, and is calculated as follows: , The baseline value is set; stress gradient variance Var (▽) σ The larger the number of ) μ It increases linearly. This factor takes into account the synergy of multiple physics fields in the battery to ensure that voltage changes are synchronized with stress changes.

[0086] For the SOH estimation module (loop level), the loss function is designed as follows:

[0087] ;

[0088] in, Based on the aging error term, SOH pred Represents the estimated SOH value. SOH true This represents the true SOH value. As the basic error term for SOH estimation, this term can adapt to the characteristics of slow battery aging, directly minimize the absolute error of SOH estimation, and avoid interference from outliers in a single cycle with the long-term degradation trend.

[0089] This is a stress-internal resistance correlation term, designed based on the strong correlation between stress difference and internal resistance in each cycle. This represents the ratio of the stress difference amplitude between the nth cycle and the first cycle. This represents the ratio of the internal resistance of the nth cycle to that of the 1st cycle. λ The association weight, with a default value of 0.8, is calculated as follows: , This is the set baseline value. This item is an aging consistency constraint, which can force the increase of stress difference amplitude to be synchronized with the increase of internal resistance, and strengthen the correlation between stress and chemical aging in the later stage of battery aging.

[0090] This is a voltage relaxation-aging coupling term. V relax This represents the static voltage drop. This represents the SOH decay rate (% / cycle). η This is the coupling coefficient, with a default value of 0.05, calculated as follows: , As set as the baseline value, when the static voltage drop deviates from the normal value by 0.05V, η The voltage increases in a stepwise manner. This design aims to force the state of oxygen (SOH) to decay more rapidly as voltage relaxation increases, which aligns with the electrochemical aging characteristics of the battery.

[0091] In practical operation, cycle charge-discharge experiments were conducted on pouch lithium-ion batteries, and distributed measurements of stress changes in different areas of the battery were performed using thin-film sensors. A schematic diagram of the distributed stress-temperature integrated acquisition device is shown below. Figure 4 As shown, the battery is fixed by a clamp so that the thin-film sensor is in close contact with the battery surface, so as to comprehensively collect the stress and temperature distribution characteristics of the battery during the charging and discharging process. The load cell is used to measure and display the initial preload applied to the battery, and the data acquisition unit collects and outputs the battery surface stress data measured by the thin-film sensor.

[0092] The experiment employed a zoned measurement strategy. This experiment collected zoned information on battery stress and temperature, as well as changes in battery stress during the charging process. Figure 5 As shown, it includes: the positive electrode tab region A, the middle region B, and the negative electrode tab region C. (The text abruptly ends here.) Figure 5 It can be intuitively seen that during battery charging and discharging, the stress value in region B, near the negative electrode tab, is the highest and changes the most significantly, while the stress value in region A, near the positive electrode tab, is the lowest and changes the least. This is because the expansion coefficient of the graphite negative electrode is much greater than that of the positive electrode material, resulting in a much greater contraction and expansion amplitude of the negative electrode during charging and discharging. Therefore, the battery as a whole and its various regions still exhibit a trend of increasing stress during charging and decreasing stress during discharging. Furthermore, since the tab region is the main current convergence point, region C, near the negative electrode tab, has a higher current density and a faster lithium-ion insertion rate, leading to a much higher expansion and contraction rate in this region compared to other regions, resulting in the most significant stress change and the highest stress value. Although the current density near the positive electrode tab is also high, the overall stress value and its variation amplitude in region A are smaller because the volume change of the positive electrode material is much smaller than that of the negative electrode. For the middle region B, the boundary constraint effect of the aluminum-plastic film encapsulation of the pouch battery has a relatively small impact on the central region; therefore, the expansion stress and its variation amplitude in region B are between those in regions A and B.

[0093] Average temperature variation in different areas of the battery, as follows Figure 6As shown, the overall battery temperature gradually increases during charging and discharging, with the temperature in region C near the negative electrode being slightly higher than in other regions. This is because ohmic heat, reaction heat, and side reaction heat are generated inside the battery during charging and discharging, leading to an increase in the overall battery temperature. Furthermore, the current density is more concentrated and the side reaction heat is more significant near the negative electrode, resulting in a higher and more drastic temperature change in the area near the negative electrode.

[0094] Figure 7 This study illustrates the change in the average stress difference Δσ (σ(region C) - σ(region A)) between the positive and negative electrode tab regions during battery cycling. It shows that the stress difference between the positive and negative electrode tab regions exhibits a clear regularity during charging and discharging, gradually increasing with charging and decreasing with discharging, consistent with the battery's SOC (State of Charge) trend and showing a strong correlation. Further investigation reveals that with battery cycling and aging, Δσ at full charge / full discharge, i.e., the maximum / minimum values ​​of Δσ, gradually increases. This is mainly attributed to the asymmetric expansion effect in the positive and negative electrode tab regions caused by the accumulation of the SEI film and lithium deposition on the negative electrode. In summary, the tab region, due to its concentrated current and heat flow characteristics, is the most stress-sensitive location in the battery structure. The stress difference between the positive and negative electrode tabs can serve as a key indicator characterizing the non-uniformity of internal reactions and the degree of structural degradation, helping to estimate the battery state and providing timely battery safety warnings.

[0095] By quantitatively analyzing the correlation between Δσ and battery state, the contribution of extracted features to SOC and SOH estimation can be effectively explored. Since stress and temperature data exhibit nonlinear trends, Spearman correlation analysis is used to comprehensively analyze the correlation between features and SOC and SOH. This method is robust to noise and suitable for nonlinear data; the calculation formula is as follows:

[0096] ;

[0097] In the formula, R(x) i ) and R(y i ) represent samples x respectively i and y i The rank of each variable.

[0098] The correlations between Δσ and the average temperature of region C during the charging process and SOC, and between the maximum and minimum values ​​of Δσ and SOH during each cycle, are calculated and shown in Table 1, where "-" represents a negative correlation. Furthermore, it can be intuitively seen that the extracted features all exhibit strong correlations with the battery state, which can help achieve accurate joint estimation of the battery state.

[0099] Table 1. Correlation between extracted features and battery state

[0100]

[0101] Then, build as follows Figure 3 The joint estimation model shown is trained. The depth (d_model) of both Transformer models is set to 64, the number of multi-head attention (num_heads) is set to 4, the hidden layer dimension in the feedforward network is set to 128, and the number of iterations is set to 200.

[0102] The real-time collected regional feature data was input into the trained model. The SOC and SOH estimation results and errors in this experiment are as follows: Figure 8 As shown, the selected features, obtained through the constructed joint estimation model, can more comprehensively characterize the internal state of the battery, enabling the model to accurately capture the dynamic trends of both SOC and SOH simultaneously. The overall error is less than 1.5%, demonstrating accurate estimation and strong robustness. In summary, the method proposed in this invention, based on distributed stress and temperature features, can achieve accurate joint estimation of SOC and SOH, and the feature data acquisition is simple, possessing certain engineering applicability.

[0103] In practical applications, stress and temperature acquisition areas can be rationally divided according to different battery types, sizes, and structures. The stress difference between the positive and negative electrode tab regions can be calculated, and appropriate Transformer model parameters and weights of various loss functions can be set according to feature data and specific requirements, thereby achieving accurate real-time estimation of SOC and SOH. This method considers the non-uniformity of battery temperature and stress distribution, and for the first time analyzes and introduces the stress difference and local temperature in the positive and negative electrode tab regions as key features. The effectiveness of the features is demonstrated through correlation analysis, proving that it can adapt to different types and structures of lithium-ion batteries. Furthermore, this method builds a joint SOC-SOH estimation model based on Transformer networks at different time scales, and constructs a multi-physics constraint design loss function using easily obtainable indicators such as stress, temperature, and voltage. The electrochemical-mechanical coupling mechanism of the battery is embedded into the deep learning model, which can adapt to the short-term dynamic behavior of SOC and the long-term degradation trend of SOH while ensuring computational efficiency, further improving the interpretability, estimation accuracy, and robustness of the model.

[0104] Example 2

[0105] A joint estimation system for the state of charge and state of health of a lithium-ion battery includes:

[0106] The data acquisition module is configured to acquire the capacity of the battery during the charge-discharge cycle, as well as the stress and temperature data at different locations, recorded in the lithium-ion battery charge-discharge experiment.

[0107] The preprocessing module is configured to preprocess the acquired data;

[0108] The correlation analysis module is configured to calculate and analyze the Spearman-level correlation coefficients between each pre-processed stress and temperature point and the state of charge and health, and to select stress and temperature features that change beyond a set degree and are correlated with the battery state beyond a set value as key features.

[0109] The model training module is configured to train the joint estimation model by taking the selected key features as input and the state of charge and health values ​​as output.

[0110] The estimation module is configured to input key features of real-time acquired stress and temperature data into the trained joint estimation model to obtain the final estimates of the state of charge and health.

[0111] The data acquisition module includes a fixing clamp, a thin-film sensor, a load cell, and a heat insulation pad. The thin-film sensor is laid on the heat insulation pad, and a lithium-ion battery is placed on the thin-film sensor. The fixing clamp clamps the three together, so that the thin-film sensor is in close contact with the surface of the lithium-ion battery. The load cell is used to collect the preload applied to the lithium-ion battery. The thin-film sensor detects the stress and temperature data at various locations on the lithium-ion battery.

[0112] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of one or more computer-usable storage media (including, but not limited to, disk storage, etc.) containing computer-usable program code. CD - ROM It takes the form of a computer program product implemented on (such as optical memory, etc.).

[0113] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0114] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0115] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0116] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for combined state-of-charge and state-of-health estimation of a lithium-ion battery, characterized in that, The method comprises the following steps: obtaining the capacity of the battery in the charging and discharging cycle and the stress and temperature data at different positions recorded in the lithium ion battery charging and discharging experiment; preprocessing the obtained data; calculating and analyzing the Spearman rank correlation coefficients of each stress and temperature point after preprocessing and the state of charge and the state of health respectively, and screening out the stress and temperature characteristics that change more than a set degree and are associated with the battery state more than a set value as key characteristics; setting a model loss function and training a joint estimation model by taking the screened key characteristics as input and the state of charge value and the state of health value as output; inputting the key characteristics of the real-time obtained stress and temperature data into the trained joint estimation model to obtain the final state of charge and state of health estimation values; the process of calculating and analyzing the Spearman rank correlation coefficients of each stress and temperature point after preprocessing and the state of charge and the state of health respectively, and screening out the stress and temperature characteristics that change more than a set degree and are associated with the battery state more than a set value as key characteristics comprises: calculating the correlation of each regional stress difference, temperature and state of charge and state of health, taking the average temperature data of each position region within the set range from the negative electrode tab region and the stress difference between the positive and negative electrode tabs as the key characteristics for estimating the state of charge, and taking the maximum and minimum values of the stress difference between the positive and negative electrode tabs as the key characteristics for estimating the state of health; the joint estimation model comprises two Transformer sub-models, for the Transformer sub-model for estimating the state of charge, the coupling relationship between the state of health is retained, the input characteristics are the stress difference Δσ between the positive and negative electrode tabs, the average temperature data of the negative electrode tab region and the state of health data collected, and the output is the state of charge estimation value; for the Transformer sub-model for estimating the state of health, the input characteristics are the maximum and minimum values of the stress difference Δσ between the positive and negative electrode tabs in each cycle, and the output characteristics are the state of health estimation value; the process of setting the model loss function comprises: for SOC estimation, the loss function is designed as follows: ； wherein the meanings of the variables are as follows: is a basis mean square error term, SOC pred represents an SOC estimation value, SOC true represents an SOC true value, α is a weight coefficient of the term, selecting the mean square error as the basis error term of the SOC estimation can accurately and sensitively capture the instantaneous fluctuation of the SOC; β • ReLU(Δ σ - σ th ) is an exponential stress risk penalty term, Δ σ represents the stress difference between the positive and negative electrode tab regions, σ th represents the stress safety threshold, which is set according to the yield strength of the specific battery material, ReLU represents a linear threshold function, β is the weight coefficient of this term, which is adjusted in the form of , is a set reference value, which is used to increase the weight and increase the penalty when Δ σ is too large, the role of this term is to activate the penalty when the stress difference exceeds the safety threshold, forcing the model to focus on high stress risk; δ •∣▽ T ∣ is the thermal gradient uniform term, ▽ T represents the temperature difference between the tab area and the center area, the degree of non-uniformity of the reaction temperature field, δ is the weight coefficient of the term, and the calculation method is , T ref is the set temperature at the beginning of the cycle, and the weight gradually increases as the temperature rises, is the set reference value, and the role of this term is to minimize the temperature distribution difference and inhibit the risk of local thermal runaway; For the electro-mechanical response constraint term, this term is designed based on the observation that the battery stress and voltage change trends are consistent and strongly correlated, represents the voltage change rate, represents the stress change rate, k is the voltage-stress coupling coefficient, μ is the coupling weight of this term, and the calculation method is , is the set reference value; the greater the stress gradient variance, μ increases linearly, and this term takes into account the multi-physical field synergy of the battery, ensuring that the voltage change and stress change are synchronized; for SOH estimation, the loss function is designed as follows: ； wherein, is a base aging error term, SOH pred represents an SOH estimate value, SOH true represents an SOH true value, which is an SOH estimation base error term, adapts the characteristics of the slow aging of the battery, minimizes the absolute error of SOH estimation, and avoids the interference of single-cycle outliers with long-period degradation trends; is the stress-internal resistance correlation term, which is designed based on the strong correlation between the stress difference and the internal resistance in each cycle, represents the ratio of the stress difference amplitude between the nth cycle and the first cycle, represents the ratio of the internal resistance between the nth cycle and the first cycle; λ is the correlation weight, which is calculated as , is the set reference value, and this term is the aging consistency constraint, which can force the stress difference amplitude growth to be synchronized with the internal resistance growth and make the correlation between the stress and the chemical aging in the later stage of battery aging be strengthened; for voltage relaxation-aging coupling term, V relax representing the resting voltage drop, representing the SOH decay rate, η for the coupling coefficient of this term, calculated as , for the set reference value, when the resting voltage drop deviates from the normal value by the set value, η increases in steps, and this term is designed to force the SOH to accelerate the decay when the voltage relaxation increases, in line with the electrochemical aging characteristics of the battery.

2. The lithium-ion battery state-of-charge and state-of-health combined estimation method of claim 1, wherein the stress The data includes the positive electrode tab region, the negative electrode tab region and the intermediate region between the two tab regions, and the intermediate region is divided into multiple regional partitions.

3. The method for jointly estimating the state of charge and state of health of a lithium-ion battery as described in claim 1, characterized in that, The temperature data includes the positive electrode tab region, the negative electrode tab region and the intermediate region between the two tab regions, and the intermediate region is divided into multiple regional partitions.

4. The method of claim 1, wherein the method further comprises: determining a state of health (SOH) of the lithium-ion battery based on the battery model and the battery parameters; and determining a state of charge (SOC) of the lithium-ion battery based on the battery model and the battery parameters. Each Transformer sub-model comprises an input layer, a feature capturing module and an output layer, wherein the input layer is used to obtain the input key characteristic data, and position encoding is performed on the key characteristic data to combine sequence information and input data; the feature capturing module is used to extract the global dependency relationship between each feature point and other points in the sequence and introduce nonlinear feature transformation; the output layer is used to integrate the dimension of the output of the feature capturing module into the target value required, and output the state of charge estimation result.

5. The method for jointly estimating the state of charge and state of health of a lithium-ion battery as described in claim 4, characterized in that, The feature capturing module includes a multi-layer stacked encoder, each encoder including a multi-head attention mechanism layer, a summation and normalization layer, a feedforward fully connected network, and a summation and normalization layer, wherein the multi-head attention mechanism layer is used to extract the global dependency between each feature point and other points in the sequence, so that each position can interact with all other positions and be dynamically weighted according to the attention score; the summation and normalization layer is used to prevent deep network gradient vanishing, accelerate convergence and improve model stability; the feedforward fully connected network includes two layers of fully connected neural networks, which are used to independently introduce nonlinear feature transformation for each position, and enhance the model expression ability.

6. A lithium-ion battery state-of-charge and state-of-health combined estimation system, applying the lithium-ion battery state-of-charge and state-of-health combined estimation method of claim 1, characterized in that, Comprise: A data acquisition module configured to acquire the capacity of the battery during the charging and discharging cycle and the stress and temperature data of different positions recorded in the lithium ion battery charging and discharging experiment; A preprocessing module configured to preprocess the acquired data; A correlation analysis module configured to calculate and analyze the Spearman rank correlation coefficients of each stress, temperature point and state of charge and state of health respectively, and screen out stress and temperature features with a change exceeding a set degree and a correlation degree with the battery state exceeding a set value as key features; A model training module configured to set the screened key features as input, the state of charge and state of health values as output, set a model loss function, and train a joint estimation model; An estimation module configured to input the key features of the real-time acquired stress and temperature data into the trained joint estimation model to obtain the final state of charge and state of health estimation values.

7. The lithium-ion battery state-of-charge and state-of-health co-estimation system of claim 6, wherein, The data acquisition module includes a fixed clamp plate, a film sensor, a load cell and a heat insulation pad, the film sensor is laid on the heat insulation pad, the lithium ion battery is arranged on the film sensor, the fixed clamp plate clamps the three, so that the film sensor is tightly attached to the surface of the lithium ion battery, the load cell is used to collect the pre-tightening force applied to the lithium ion battery, and the film sensor is used to detect the stress and temperature data of each area position of the lithium ion battery.

Images