Method for predicting SOH and RUL of lithium ion battery based on ICA-DVA analysis

Abstract

The invention discloses a method for analyzing and predicting SOH and RUL of a lithium ion battery based on ICA-DVA, and the method comprises the following steps: 1, reconstructing the voltage through a voltage reconstruction second-order RC model, and obtaining an IC curve and a DV curve; 2, feature extraction: extracting features representing battery aging from the IC and DV curves; thirdly, constructing a feature prediction model, an SOH prediction model and an RUL prediction model, and inputting the features representing the aging of the battery extracted in the second step into the SOH prediction model to obtain a predicted value of the SOH; and fourthly, decomposing the SOH prediction result, and performing RUL prediction by using the decomposed result. According to the method, high-precision SOH and RUL prediction of the lithium ion battery is realized by combining the second-order RC model, ICA / DVA analysis, the BiGRU-Attention model and the CEEMDAN algorithm. According to the method, prediction precision and robustness are improved, adaptability to battery inconsistency and temperature uncertainty is enhanced, and reliable technical support is provided for safe and efficient operation of an energy storage system.

Description

Technical Field

[0001] The present invention relates to the field of energy storage systems, and specifically to a method for predicting the state of health (SOH) and remaining useful life (RUL) of lithium-ion batteries based on ICA / DVA analysis. Background Art

[0002] With the continuous growth of global energy demand and the increasingly serious environmental pollution problems, the development of renewable energy has become a key direction to ensure energy security and achieve sustainable development. However, clean energies such as wind energy and photovoltaic energy have inherent defects in power supply intermittency, making it difficult for them to be stably connected to the grid. This challenge has promoted the rapid development of energy storage technologies, among which electrochemical energy storage has attracted much attention due to its high efficiency and flexibility. Among various electrochemical energy storage technologies, lithium-ion batteries have become the mainstream energy storage solution due to their advantages such as high energy density, long cycle life, and strong environmental adaptability, and are widely used in fields such as power grid peak shaving and power quality optimization.

[0003] As the core guarantee technology for the safe operation of lithium-ion batteries, the battery management system (BMS) realizes the accurate estimation of the state of charge (SOC) and state of health (SOH) of the battery and the prediction of the remaining useful life (RUL) by real-time monitoring of key parameters such as voltage, current, and temperature. Among them, SOH is used to characterize the attenuation degree of the current maximum capacity of the battery relative to the initial capacity, and RUL is used to predict the remaining available cycle times before the battery fails. Research shows that accurately predicting SOH and RUL can not only timely warn of the battery failure risk, but also extend the battery life by more than 30% through optimizing the charge and discharge strategy, thus significantly reducing the full life cycle cost of the energy storage system.

[0004] Nevertheless, lithium-ion batteries still face severe challenges in large-scale energy storage applications: overcharging or over-discharging is likely to cause thermal runaway accidents, the sudden drop in capacity caused by cyclic attenuation will reduce the system reliability, and potential fault hazards may also endanger personal and property safety. Therefore, current research is committed to integrating electrochemical models with machine learning algorithms to provide technical support for building a safe and efficient clean energy system by improving the prediction accuracy and working condition adaptability of SOH and RUL.

[0005] The technical idea of the present invention is: to perform precise voltage reconstruction, mine hidden battery information, generate IC (Incremental Capacity) and DV (Differential Voltage) curves, extract peak height, peak position, peak area, and capacity (Q) from them as comprehensive aging characteristics, improve the reliability, effectiveness of feature extraction and the input quality of the prediction model, and build a prediction model to more accurately predict SOH and RUL. Summary of the Invention

[0006] The object of the present invention is to provide a method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis, so as to achieve high-precision prediction of the SOH and RUL of the lithium-ion battery.

[0007] To achieve the above object, a method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis according to the present invention is characterized by the following steps:

[0008] The first step is to reconstruct the voltage through a voltage reconstruction second-order RC model to obtain the IC and DV curves;

[0009] The second step is feature extraction, extracting features characterizing battery aging from the IC and DV curves;

[0010] The third step is to construct a feature prediction model, an SOH prediction model, and an RUL prediction model, input the features characterizing battery aging extracted in the second step into the SOH prediction model to obtain the predicted value of SOH;

[0011] The fourth step is to decompose the SOH prediction result and use the decomposed result for RUL prediction.

[0012] In the first step, in order to establish the voltage reconstruction second-order RC model, first use the second-order equivalent circuit model to simulate the dynamic behavior of the lithium-ion battery. The formula of the second-order equivalent circuit model is:

[0013]

[0014] The parameters in the second-order equivalent circuit model are respectively described as follows:

[0015] Terminal voltage U, unit is volt;

[0016] State of charge SOC, as a percentage value, obtained through the SOC calculation formula. The specific meaning is the percentage of the remaining battery capacity to the maximum available capacity;

[0017] Open-circuit voltage U ocv (SOC), unit is volt, U ocv (SOC) is obtained through the open-circuit voltage calculation formula;

[0018] Ohmic internal resistance R0, unit is ohm, obtained through experimental measurement or parameter identification, characterizing the inherent resistance of the battery material;

[0019] Polarization internal resistance R p1 , R p2 , unit is ohm;

[0020] Q is the charging capacity; the unit is ampere-hour, a measured value, obtained through the battery management system BMS;

[0021] Time constants τ1 and τ2, in seconds;

[0022] Current I, in amperes, measured value;

[0023] Polarization voltage U p1 and U p2 , in volts;

[0024] Polarization internal resistance R p1 and R p2 、Time constants τ1 and τ2 and polarization voltage U p1 and U p2 are all determined through dynamic parameter identification, where parameter identification refers to determining the polarization internal resistance, time constants, and polarization voltage in the model through experimental data.

[0025] The open-circuit voltage calculation formula is:

[0026]

[0027] where E is the activation energy, in joules per mole, sourced from experimental calibration or battery material characteristics;

[0028] Gas constant R, in joules per mole per kelvin, a physical constant, R = 8.314;

[0029] Reference temperature T ref , in kelvin or degrees Celsius; experimental set value, usually taken as 25°C, i.e., 298.15K;

[0030] Current temperature T, in kelvin or degrees Celsius, a real-time measured value;

[0031] a i is a dimensionless constant, a fitted value;

[0032] m is the order of the polynomial in SOC.

[0033] The SOC calculation formula is:

[0034]

[0035] where S0 is the initial SOC at the start of charging, a dimensionless percentage, estimated in real time by the battery management system or experimentally calibrated;

[0036] Q is the charging capacity, in ampere-hours, a real-time measured value;

[0037] Q max is the current maximum capacity, in ampere-hours, obtained through experimental testing.

[0038] In the first step, by combining the SOC calculation formula, the open-circuit voltage calculation formula, and the formula of the second-order equivalent circuit model, a second-order RC model for voltage reconstruction is obtained; the formula of the second-order RC model for voltage reconstruction is:

[0039]

[0040] The second-order RC model for voltage reconstruction is used to fit the Q-V curve to achieve voltage reconstruction.

[0041] Among them, each parameter is the same as the homonymous parameter in the SOC calculation formula, the open-circuit voltage calculation formula, and the formula of the second-order equivalent circuit model.

[0042] In the second step, the features characterizing battery aging are extracted from the IC and DV curves, including peak height, peak position, peak area, and Q.

[0043] In the third step, a feature prediction model, an SOH prediction model, and an RUL prediction model are constructed using the BiGRU-Attention model.

[0044] First, the peak height, peak position, peak area, and Q are respectively predicted through the feature prediction model to obtain the predicted values of the peak height, peak position, peak area, and Q. Then, the predicted values of the peak height, peak position, peak area, and Q are input into the SOH prediction model to obtain the predicted value of SOH.

[0045] The CEEMDAN algorithm is used to decompose the SOH sequence, and the residual sequence is used as the input of the RUL prediction model to achieve the prediction of RUL, and the uncertainty of the RUL prediction result is analyzed in combination with the Monte Carlo algorithm.

[0046] The present invention has the following advantages:

[0047] 1. The prediction accuracy is improved through aspects such as voltage reconstruction, feature extraction, and prediction models.

[0048] By reconstructing the voltage through the second-order RC model, the dynamic behavior of the battery can be accurately simulated, the noise influence can be reduced, and the accuracy of feature extraction can be improved.

[0049] By reconstructing the voltage through the second-order RC model, the dynamic behavior of the battery can be accurately simulated, the noise influence can be reduced, and the accuracy of feature extraction can be improved.

[0050] Through the BiGRU-Attention prediction model, the non-linear and non-stationary SOH degradation curve can be effectively processed, and the prediction accuracy and robustness can be improved.

[0051] 2. The robustness and adaptability are improved.

[0052] The present invention is applicable to single batteries and battery packs, capable of handling battery inconsistency issues and ensuring prediction performance under complex operating conditions. By combining the Arrhenius law and considering the influence of temperature on battery characteristics, the adaptability of the model under different temperature conditions is enhanced.

[0053] 3. Feature extraction and prediction.

[0054] Through comparison with the Pearson correlation coefficient, it is verified that the features extracted by the proposed method are more accurate than traditional methods (such as Gaussian filtering and moving average filtering). The BiGRU-Attention model can capture the mapping relationship between features and SOH, and the prediction curve is closer to the real curve with smaller errors.

[0055] The SOH sequence is decomposed by the CEEMDAN algorithm, and the residual sequence is used as the input of the RUL prediction model, effectively avoiding the influence of capacity regeneration phenomenon and improving the accuracy and robustness of RUL prediction.

[0056] The second-order equivalent circuit model for simulating the dynamic behavior of lithium-ion batteries proposed by the present invention and the voltage reconstruction second-order RC model constructed on this basis are not only applicable to single batteries but also can well handle the problem of battery inconsistency.

[0057] In terms of model structure design:

[0058] The dynamic behavior of the battery is simulated through two polarization resistors (Rp1, Rp2) and two polarization capacitors (Cp1, Cp2). This design can more accurately capture the voltage change characteristics of the battery during charge and discharge.

[0059] Multi-time constant characteristics: Two time constants (τ1, τ2) can simulate the fast and slow dynamic responses of the battery respectively. This multi-time constant characteristic enables the model to adapt to the aging characteristics and dynamic behaviors of different batteries.

[0060] In terms of parameter identification and self-adaptability:

[0061] Parameter identification: The parameters of the model are identified by the nonlinear least squares optimization method. These parameters include polarization voltages (Up1, Up2), polarization internal resistances (Rp1, Rp2), ohmic internal resistance (R0), etc.

[0062] Adaptive update: The model parameters will be continuously updated as the battery ages, ensuring that the model can dynamically adapt to the changes in battery performance. This self-adaptability makes the model applicable not only to single batteries but also to handle the inconsistency of different batteries in a battery pack.

[0063] In terms of handling battery inconsistency:

[0064] Adapting to individual differences: Each battery in a battery pack may have differences in capacity, internal resistance, self-discharge rate, etc. The second-order RC model can adapt to the characteristics of each battery individually through parameter identification and adaptive updating, thereby handling battery inconsistencies.

[0065] Robustness: The model is highly robust to noise and measurement errors, maintaining stable performance under complex operating conditions. This robustness makes the model more reliable when handling battery inconsistencies.

[0066] In terms of fitting accuracy:

[0067] High fitting accuracy: By combining the Arrhenius law, the voltage reconstruction second-order RC model can accurately fit the QV curve, reshape the IC / DV curve, mine implicit battery information, and reduce the impact of noise. The present invention extracts peak height, peak position, peak area, and Q from the IC and DV curves as a feature combination to characterize battery aging. These features can better reflect the aging details of the battery than the features extracted by traditional methods. Therefore, these features can not only more accurately assess the current health status of the battery, but also use these extracted features as input to train the BiGRU-Attention prediction model, which can more accurately predict the battery's SOH and RUL.

[0068] Accurate fitting of voltage platform: The voltage reconstruction second-order RC model can more accurately fit the voltage platform of the battery, providing a high-quality data foundation for subsequent feature extraction and prediction.

[0069] Accurate voltage reconstruction is the basis for generating high-fidelity IC / DV curves and directly affects the reliability of extracting aging characteristics (peak height, peak position, etc.).

[0070] These characteristics make the voltage reconstruction second-order RC model not only suitable for single batteries, but also can handle the inconsistency problem in battery packs well.

[0071] In summary, this paper achieves high-precision prediction of lithium-ion battery SOH and RUL by combining a second-order RC model, ICA / DVA analysis, the BiGRU-Attention model, and the CEEMDAN algorithm. This method not only improves prediction accuracy and robustness, but also enhances adaptability to battery inconsistencies and temperature uncertainties, providing reliable technical support for the safe and efficient operation of energy storage systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 Schematic diagram of the framework of the lithium battery SOH and RUL prediction method based on ICA / DCA analysis.

[0073] Figure 2 Schematic diagram of the second-order RC equivalent circuit model.

[0074] Figure 3 It is a schematic diagram of voltage reconstruction based on the second-order RC model.

[0075] Figure 4 It is a schematic diagram of feature extraction and prediction.

[0076] Figure 5 It is a schematic diagram of SOH and RUL prediction. Specific implementation manners

[0077] As Figures 1 to 5 shown, the present invention discloses a method for predicting SOH and RUL of lithium-ion batteries based on ICA-DVA analysis.

[0078] A method for predicting SOH and RUL of lithium-ion batteries based on ICA-DVA analysis, which is characterized by the following steps:

[0079] The first step is to reconstruct the voltage through a voltage reconstruction second-order RC model to obtain IC (Incremental Capacity) and DV (Differential Voltage) curves;

[0080] The second step is feature extraction, extracting features characterizing battery aging from the IC and DV curves;

[0081] The third step is to construct a feature prediction model, an SOH prediction model and an RUL prediction model, input the features characterizing battery aging extracted in the second step into the SOH prediction model to obtain the predicted value of SOH;

[0082] The fourth step is to decompose the SOH prediction result and use the decomposed result for RUL prediction.

[0083] The present invention has the following advantages:

[0084] 1. Improve the prediction accuracy through aspects such as voltage reconstruction, feature extraction and prediction models.

[0085] By reconstructing the voltage through the second-order RC model, the dynamic behavior of the battery can be accurately simulated, the influence of noise can be reduced, and the accuracy of feature extraction can be improved.

[0086] By reconstructing the voltage through the second-order RC model, the dynamic behavior of the battery can be accurately simulated, the influence of noise can be reduced, and the accuracy of feature extraction can be improved.

[0087] Through the BiGRU-Attention prediction model, the non-linear and non-stationary SOH degradation curve can be effectively processed, and the prediction accuracy and robustness can be improved.

[0088] 2. Improve the robustness and adaptability.

[0089] The present invention is applicable to single batteries and battery packs, capable of handling battery inconsistency issues and ensuring prediction performance under complex working conditions. By combining the Arrhenius law and considering the influence of temperature on battery characteristics, the adaptability of the model under different temperature conditions is enhanced.

[0090] 3. Feature extraction and prediction.

[0091] Through the comparison of Pearson correlation coefficients, it is verified that the features extracted by the proposed method are more accurate than traditional methods (such as Gaussian filtering and moving average filtering). The BiGRU-Attention model can capture the mapping relationship between features and SOH, and the prediction curve is closer to the real curve with smaller errors.

[0092] The SOH sequence is decomposed by the CEEMDAN algorithm, and the residual sequence is used as the input of the RUL prediction model, effectively avoiding the influence of capacity regeneration phenomenon and improving the accuracy and robustness of RUL prediction.

[0093] In summary, the present invention realizes high-precision prediction of SOH and RUL of lithium-ion batteries by combining the second-order RC model, ICA / DVA analysis, BiGRU-Attention model, and CEEMDAN algorithm. This method not only improves the prediction accuracy and robustness but also enhances the adaptability to battery inconsistency and temperature uncertainty, providing reliable technical support for the safe and efficient operation of energy storage systems.

[0094] In the first step, in order to establish the voltage reconstruction second-order RC model, first, the dynamic behavior of lithium-ion batteries is simulated with a second-order equivalent circuit model. The formula of the second-order equivalent circuit model is:

[0095]

[0096] The parameters in the second-order equivalent circuit model are described as follows:

[0097] The terminal voltage U, with the unit of volt;

[0098] The state of charge SOC, as a percentage value, obtained through the SOC calculation formula. Its specific meaning is the percentage of the remaining battery capacity to the maximum available capacity;

[0099] The open-circuit voltage U ocv (SOC), with the unit of volt. U ocv (SOC) is obtained through the open-circuit voltage calculation formula;

[0100] The ohmic internal resistance R0, with the unit of ohm, obtained through experimental measurement or parameter identification, representing the inherent resistance of the battery material;

[0101] The polarization internal resistance R p1 , Rp2 , in ohms;

[0102] Q is the charging capacity; the unit is ampere-hour (Ah), a measured value obtained through the battery management system BMS;

[0103] The time constants τ1 and τ2, in seconds;

[0104] The current I, in amperes, a measured value;

[0105] The polarization voltage U p1 and U p2 , in volts.

[0106] The polarization internal resistance R p1 and R p2 、the time constants τ1 and τ2, and the polarization voltage U p1 and U p2 are all determined through dynamic parameter identification. Here, parameter identification refers to determining the polarization internal resistance, time constant, and polarization voltage in the model through experimental data (such as charge / discharge voltage / current curves).

[0107] The technical consideration for first using a second-order equivalent circuit model to simulate the dynamic behavior of a lithium-ion battery is that: as Figure 2 and Figure 3 shown, since the terminal voltage of a lithium battery during charge and discharge is determined by the motor voltage and resistance, and the terminal voltage curve contains the information in the motor voltage curve, in order to avoid the complex parameter adjustment problem in traditional filtering methods, a second-order equivalent circuit model is first used to simulate the dynamic behavior of a lithium-ion battery.

[0108] The open-circuit voltage calculation formula is:

[0109]

[0110] The open-circuit voltage calculation formula introduces the Arrhennius law, considering the influence of temperature on the function.

[0111] where E is the activation energy, in joules per mole (J / mol), obtained from experimental calibration (such as electrochemical impedance spectroscopy, charge / discharge tests) or battery material characteristics (such as literature values of the activation energy of the positive and negative electrodes);

[0112] The gas constant R, in joules per mole per kelvin (J / (mol·K)), is a physical constant, R = 8.314;

[0113] The reference temperature T ref , in kelvin (K) or degrees Celsius (°C); an experimental set value, usually taken as 25°C, i.e., 298.15K;

[0114] The current temperature T, in Kelvin (K) or Celsius (°C), is a real-time measurement value (e.g., obtained through a temperature sensor);

[0115] a i is a dimensionless constant, a fitted value;

[0116] m is the order of the polynomial in the SOC.

[0117] The open-circuit voltage calculation formula has the following technical considerations: The Uocr(SOC) function depends not only on the degree of battery aging but also on temperature changes. Therefore, the Arrhenius law is introduced in the present invention to consider the influence of temperature on the Uocr(SOC) function.

[0118] The SOC calculation formula is:

[0119]

[0120] where S0 is the initial SOC at the start of charging, a dimensionless percentage, estimated in real time through the battery management system (BMS) or experimentally calibrated (the initial state of charge at the start of charging);

[0121] Q is the charging capacity, in ampere-hours (Ah), a real-time measurement value (Coulomb counting method, i.e., the integral of current over time);

[0122] Q max is the current maximum capacity, in ampere-hours (Ah), obtained through experimental testing (the maximum available capacity under the current battery aging state, which decays with the number of cycles).

[0123] In the first step, combining the Arrhenius law, the SOC calculation formula, the open-circuit voltage calculation formula, and the formula of the second-order equivalent circuit model, a voltage reconstruction second-order RC model is obtained; the formula of the voltage reconstruction second-order RC model is:

[0124]

[0125] The voltage reconstruction second-order RC model is used to fit the Q-V curve to achieve voltage reconstruction,

[0126] where the parameters are the same as the homonymous parameters in the SOC calculation formula, the open-circuit voltage calculation formula, and the formula of the second-order equivalent circuit model.

[0127] The formula of the voltage reconstruction second-order RC model can be simplified to:

[0128]

[0129] Among them, V represents the terminal voltage; Q is the remaining capacity; b1 represents the Arrhenius term in the equation, which is the only constant related to temperature and should be in the range of 0.5 - 1.5, including both ends; Q ∈ [Q end , Q max,0 ; where Q end is the charging capacity at the end of the CC charging (i.e., constant current charging) process, and Q max,0 is the initial maximum capacity of the battery; S0 represents the initial SOC when starting charging, so Q max should be restricted between [0, 1] (including both ends); a i represents the constant term in the U ocv (SOC) function,

[0130] a i ∈ (-∞, +∞); b1, b2, b3, b4, b5, b6 respectively represent the abbreviations of the corresponding terms and should be in the range of (-∞, 0]. The parameters in this formula are obtained by fitting the Q - V curve, and the parameter identification problem is attributed to the nonlinear least - squares optimization problem. The constant a is obtained by fitting the initial Q - V curve; b1, b2, b3, b4, b5, b6, S0 and Q i are continuously updated as the battery ages. max In the second step, as

[0131] shown, based on the reconstructed voltage results, the IC and DV curves are reconstructed, and the features characterizing battery aging are extracted from the IC and DV curves, including taking the peak height, peak position, peak area and Q (i.e., charging capacity). The feature extraction results of the method proposed in this paper are compared with those of the Gaussian filtering and moving average filtering methods through the Pearson correlation coefficient. Figure 4 The ICA and DVA methods can not only analyze the battery aging mechanism by using the changes of their peaks and valleys, but also extract battery aging information from the IC and DV curves as health indicators for SOH and RUL prediction.

[0132] In the third step, the BiGRU - Attention model is used to construct a feature prediction model, an SOH prediction model and an RUL prediction model.

[0133] First, the peak height, peak position, peak area and Q are predicted respectively through the feature prediction model to obtain the predicted values of the peak height, peak position, peak area and Q, and then the predicted values of the peak height, peak position, peak area and Q are input into the SOH prediction model to obtain the predicted value of SOH.

[0134] Finally, the predicted value of SOH is input into the RUL prediction model to obtain the predicted value of RUL.

[0135] The BiGRU-Attention model is a deep learning architecture that combines bidirectional gated recurrent units (BiGRU) and the attention mechanism (Attention). It is widely used in fields such as text sentiment analysis, time series prediction, and relation extraction. Its core idea is to capture the forward and backward dependencies of the sequence through a bidirectional network and use the attention mechanism to dynamically allocate weights to enhance the representation of key features. The typical structure of the BiGRU-Attention model is: input layer → BiGRU layer → attention layer → fully connected layer → output layer. The BiGRU-Attention model is a conventional technology, and the feature prediction model, SOH prediction model, and RUL prediction model are not described in detail.

[0136] The SOH prediction model constructed in the present invention is closest to the real curve, has a smaller error than other models, and can avoid the influence of the capacity regeneration phenomenon, improving the SOH prediction accuracy.

[0137] Since the SOH degradation curve is non-linear and non-stationary, in order to avoid the influence of capacity regeneration, the CEEMDAN algorithm is used to decompose the SOH sequence, and the residual sequence is used as the input of the RUL prediction model to achieve the prediction of RUL. The Monte Carlo algorithm is combined to analyze the uncertainty of the RUL prediction results. The CEEMDAN algorithm and the Monte Carlo algorithm are both conventional technologies and are not described in detail.

[0138] The above embodiments are only used to illustrate rather than limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the present invention can still be modified or equivalently replaced, and any modification or partial replacement without departing from the spirit and scope of the present invention shall be covered by the scope of the claims of the present invention.

Claims

1. A method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis, characterized in that Proceed as follows: The first step is to reconstruct the voltage using a second-order RC model for voltage reconstruction to obtain the IC and DV curves; The second step is feature extraction, extracting features characterizing battery aging from the IC and DV curves; The third step is to construct a feature prediction model, an SOH prediction model, and an RUL prediction model. Input the features characterizing battery aging extracted in the second step into the SOH prediction model to obtain the predicted value of SOH; The fourth step is to decompose the SOH prediction result and use the decomposed result for RUL prediction.

2. A method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis according to claim 1, characterized in that: In the first step, in order to establish a second-order RC model for voltage reconstruction, first use a second-order equivalent circuit model to simulate the dynamic behavior of the lithium-ion battery. The formula for the second-order equivalent circuit model is: The parameters in the second-order equivalent circuit model are as follows: Terminal voltage U, in volts; State of charge SOC, as a percentage value, obtained through the SOC calculation formula. The specific meaning is the percentage of the remaining battery capacity to the maximum available capacity; Open-circuit voltage U ocv (SOC), in volts, U ocv (SOC) is obtained through the open-circuit voltage calculation formula; Ohmic internal resistance R0, in ohms, obtained through experimental measurement or parameter identification, characterizing the inherent resistance of the battery material; Polarization internal resistance R p1 , R p2 , in ohms; Q is the charging capacity; in ampere-hours, a measured value obtained through the battery management system BMS; Time constants τ1 and τ2, in seconds; Current I, in amperes, a measured value; Polarization voltage U p1 and U p2 , in volts; Polarization internal resistance R p1 and R p2 、time constants τ1 and τ2, and polarization voltage U p1 and U p2 are all determined through dynamic parameter identification. Here, parameter identification refers to determining the polarization internal resistance, time constants, and polarization voltage in the model through experimental data.

3. A method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis according to claim 2, characterized in that: The open-circuit voltage calculation formula is: where E is the activation energy, in joules per mole, from experimental calibration or battery material characteristics; Gas constant R, in joules per mole per kelvin, a physical constant, R = 8.314; Reference temperature T ref , in Kelvin or Celsius; experimental set value, usually taken as 25 °C i.e. 298.15 K; Current temperature T, in kelvin or degrees Celsius, a real-time measured value; a i is a dimensionless constant, the fitted value; m is the order of the polynomial in SOC.

4. A method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis according to claim 3, characterized in that: The SOC calculation formula is: where S0 is the initial SOC at the start of charging, a dimensionless percentage, estimated in real time through the battery management system or experimentally calibrated; Q is the charging capacity, in ampere-hours, a real-time measured value; Q max is the current maximum capacity, in ampere-hours, obtained through experimental testing.

5. A method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis according to claim 4, characterized in that: In the first step, combining the SOC calculation formula, the open-circuit voltage calculation formula, and the formula of the second-order equivalent circuit model, a second-order RC model for voltage reconstruction is obtained. The formula for the second-order RC model for voltage reconstruction is: The second-order RC model for voltage reconstruction is used to fit the Q-V curve to achieve voltage reconstruction, where the parameters are the same as the homonymous parameters in the SOC calculation formula, the open-circuit voltage calculation formula, and the formula of the second-order equivalent circuit model.

6. A method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis according to claim 5, characterized in that: In the second step, the features characterizing battery aging extracted from the IC and DV curves include peak height, peak position, peak area, and Q.

7. A method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis according to claim 6, characterized in that: In the third step, the BiGRU-Attention model is used to construct a feature prediction model, an SOH prediction model, and an RUL prediction model. First, the peak height, peak position, peak area, and Q are predicted respectively by the feature prediction model to obtain the predicted values of the peak height, peak position, peak area, and Q. Then, the predicted values of the peak height, peak position, peak area, and Q are input into the SOH prediction model to obtain the predicted value of the SOH.

8. A method for predicting the SOH and RUL of a lithium-ion battery based on ICA-DVA analysis according to claim 7, characterized in that: The CEEMDAN algorithm is used to decompose the SOH sequence, and the residual sequence is used as the input of the RUL prediction model to realize the prediction of the RUL, and the Monte Carlo algorithm is combined to analyze the uncertainty of the RUL prediction result.

Images