A joint estimation method for high-sensitivity parameters and SOC of lithium batteries

Abstract

This invention relates to a joint estimation method for high-sensitivity parameters and State of Charge (SOC) of lithium batteries, belonging to the field of lithium battery energy storage technology. The invention obtains measured data through dynamic operating condition experiments of lithium batteries. Based on the structural characteristics of the equivalent circuit model and the actual identifiability of the model parameters, high-sensitivity parameters are selected. Then, a parameter estimator is designed to identify only the high-sensitivity parameters, while other battery model parameters are obtained using offline parameter identification methods, determining the mapping relationship between each parameter and SOC. Finally, a state estimator with an adaptive module is designed, combining the parameter estimator and the state estimator to establish an adaptive double-extended Kalman filter algorithm, achieving joint estimation of model parameters and SOC. This invention reduces computational costs by decreasing the number of identified parameters, enables real-time updates of high-sensitivity model parameters, and possesses high SOC estimation accuracy. It also exhibits strong robustness under interference from current and voltage measurement errors, making it highly practical.

Description

Technical Field

[0001] This invention belongs to the field of lithium battery energy storage technology, and relates to a method for joint estimation of high sensitivity parameters and SOC of lithium batteries. Specifically, it relates to a method for joint estimation of high sensitivity parameters and SOC based on adaptive dual extended Kalman filtering. Background Technology

[0002] Lithium-ion batteries are widely used in electric vehicles due to their advantages such as high energy density, long cycle life, environmental friendliness, and low self-discharge rate. However, due to technological limitations, many problems remain to be solved, with inaccurate estimation of the battery's State of Charge (SOC) being one of the core issues. Accurate SOC estimation is fundamental to predicting the remaining driving range of an electric vehicle, effectively preventing breakdowns and directly impacting the user's driving experience. Furthermore, the Battery Management System (BMS) uses SOC as a crucial threshold for charge and discharge management, dynamically adjusting the battery's state of charge and discharge by monitoring SOC. This is significant for adjusting energy distribution strategies, improving battery utilization, preventing overcharging and over-discharging, and ensuring personal safety. Because a lithium battery is a complex, nonlinear, time-varying system, its SOC cannot be directly measured by sensors and must be estimated based on external state parameters such as current, voltage, and temperature. However, accurate SOC estimation is extremely difficult due to: the time-varying characteristics of the battery; the influence of factors such as temperature, aging, operating conditions, and data sampling accuracy; and the reliance on an accurate battery model, which has numerous parameters, making accurate identification a significant challenge.

[0003] Current research on SOC estimation mainly focuses on the improvement and optimization of models and algorithms. These attempts have shown initial success, indeed providing new ideas for improving the accuracy of battery SOC estimation. However, the increased complexity of models and algorithms has also significantly increased the computational cost of the controller. Given the complex and variable operating environment of lithium batteries and the limited computational capabilities of battery management systems, the anti-interference ability and practical value of these algorithms need further verification. Algorithms applicable to embedded systems must possess characteristics such as simple models, stability, reliability, and strong anti-interference capabilities. Considering that SOC estimation is primarily based on models, it is necessary to establish accurate battery models through reasonable parameter identification methods, correctly assess the factors affecting SOC estimation accuracy, balance algorithm complexity and SOC accuracy, and design a safe, reliable, robust, and computationally low SOC estimation method. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a joint estimation method for high-sensitivity parameters and State of Charge (SOC) of lithium batteries. Based on adaptive dual extended Kalman filtering, this method achieves joint estimation of high-sensitivity parameters and SOC, simultaneously balancing algorithm complexity and SOC estimation accuracy, and improving the algorithm's anti-interference capability. This invention achieves its objective through the following technical solutions.

[0005] A method for jointly estimating high-sensitivity parameters and State of Charge (SOC) of lithium batteries is characterized by the following steps: First, measured data is obtained through dynamic operating condition experiments of lithium batteries. Based on the structural characteristics of the equivalent circuit model and the actual identifiability of the model parameters, high-sensitivity parameters are selected. Then, a parameter estimator that only identifies high-sensitivity parameters is designed, while other battery model parameters are obtained using offline parameter identification methods to determine the mapping relationship between each parameter and SOC. Finally, a state estimator with an adaptive module is designed, and the parameter estimator and the state estimator are combined to establish an adaptive double extended Kalman filter algorithm to achieve joint estimation of model parameters and SOC.

[0006] Furthermore, the high-sensitivity parameters are open-circuit voltage and ohmic internal resistance.

[0007] Furthermore, the offline parameter identification method includes particle swarm optimization, least squares method, and curve fitting method.

[0008] Furthermore, the adaptive module employs Sage-Husa filtering, the parameter estimator uses the ohmic internal resistance R0 as the state variable, and the state estimator uses SOC and polarization voltage as state variables, both of which use terminal voltage as the observation.

[0009] Furthermore, a joint estimation method for high-sensitivity parameters and SOC of lithium batteries specifically includes the following steps:

[0010] S1 built an experimental platform to obtain measured data through dynamic operating condition experiments of lithium batteries. Based on the structural characteristics of the equivalent circuit model, it analyzed and compared the actual identifiability of each parameter of the model and determined the open-circuit voltage U. ocv The internal resistance in ohms is a high-sensitivity parameter;

[0011] The S2 design only identifies the open-circuit voltage U. ocv A parameter estimator for the ohmic internal resistance R0 is used to determine the open-circuit voltage U through the OCV-SOC fitting curve. ocv The mapping relationship between SOC and the optimal polynomial order varies depending on the battery type. Taking a 5th-order polynomial fitting as an example:

[0012]

[0013] Let the parameter to be identified, θ, be represented as R0 and U. ocv Combination of fitting parameters:

[0014] θ = [R0, m0, m1, m2, m3, m4, m5] T (2)

[0015] The S3 design incorporates a state estimator with an adaptive module, which, combined with a parameter estimator, establishes an Adaptive Double Extended Kalman Filter (ADEKF) algorithm to jointly estimate model parameters and state of charge (SOC). The adaptive module employs a Sage-Husa filter, the parameter estimator uses ohmic internal resistance as the state variable, and the state estimator uses SOC and polarization voltage as state variables. Both use terminal voltage as the observation. The state-space equation of the ADEKF algorithm is as follows:

[0016]

[0017] Where θ k x k r is a state variable k w k For process noise, v k e k For observation noise, process noise ω k r k and observation noise υ k e k All are Gaussian white noise with a mean of 0 and are mutually independent, with covariance matrices Q and Q', respectively. x Q θ R x and R θ .

[0018] Furthermore, the SOC estimation process based on the ADEKF algorithm in step S3 is as follows:

[0019] 1) Initialization of parameter estimator and state estimator: including x0, θ0、 and

[0020] 2) Prior estimation and prediction of parameters and states, including:

[0021] State estimator time update:

[0022]

[0023] Time updates for parameter estimators:

[0024]

[0025] 3) Posterior estimation and correction of parameters and states, including:

[0026] State estimator measurement update:

[0027]

[0028] Parameter estimator measurement update:

[0029]

[0030] 4) Update of process noise and observation noise covariance of the state estimator:

[0031]

[0032] in:

[0033]

[0034] In equations (4) to (9), those marked with "-" represent the prior estimate at the current time, and those marked with "+" represent the posterior estimate. and These represent the Kalman gains of the state estimator and the parameter estimator, respectively. and Then, these represent the noise covariance matrices of the two estimators, respectively.

[0035] The present invention has the following advantages over existing technologies:

[0036] 1. The method for joint estimation of high-sensitivity parameters and SOC provided by this invention includes an ADEKF algorithm comprising two extended Kalman filters, which serve as parameter estimators and state estimators, respectively. This allows for real-time updates of model parameters and SOC in sequence, with the two being coupled together to achieve high SOC estimation accuracy.

[0037] 2. The parameter estimator designed in this invention only identifies two high-sensitivity parameters, open-circuit voltage and ohmic internal resistance. The remaining battery model parameters are obtained using an offline parameter identification method. By reducing the number of parameters to be identified, the algorithm complexity and computational cost are reduced.

[0038] 3. The state estimator of this invention incorporates an adaptive module based on Sage-Husa filtering, which can update the covariance matrix of process noise and observation noise in real time. This solves the problem that in practice, the accuracy of SOC estimation is reduced or even fails to converge due to inaccurate noise statistics, and it is applicable to complex working conditions.

[0039] 4. The ADEKF algorithm has good robustness under the interference of current and voltage measurement errors and has great practical value. Attached Figure Description

[0040] Figure 1 This is a flowchart of the method of the present invention;

[0041] Figure 2 This is a schematic diagram of the second-order RC equivalent circuit model structure in the embodiment;

[0042] Figure 3a The SOC estimation results of the ADEKF algorithm under FUDS conditions are shown. Figure 3b This is the corresponding SOC error curve;

[0043] Figure 4a The results of SOC simulation estimation using the ADEKF algorithm under different current noise interferences are shown. Figure 4b This is the corresponding SOC error curve. Detailed Implementation

[0044] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0045] A method for evaluating the actual identifiability of parameters in an equivalent circuit model of a lithium battery, the flowchart of which is shown below. Figure 1 As shown. In this embodiment, an 18650 ternary lithium battery with a nominal capacity of 2500mAh and a nominal voltage of 3.7V is used as the experimental object. The specific implementation steps are as follows:

[0046] 1. An experimental platform was built, and experimental data was obtained through dynamic operating condition experiments commonly used in lithium battery testing, such as the Federal Urban Driving Schedule (FUDS) experiment. An appropriate equivalent circuit model was selected; considering both model complexity and accuracy, this embodiment chose... Figure 2 The second-order RC model shown, where U ocv U represents the open-circuit voltage. L Represents the terminal voltage, I represents the operating current, R0 is the internal resistance in ohms, and R p1 R p2 For polarization resistance, C p1 C p2 For polarized capacitors, the model needs to identify R0 and R... p1 R p2 C p1 C p2 and U ocv Six parameters were considered. First, the actual identifiability level of each parameter in the model was studied, and it was found that the actual identifiability of open-circuit voltage and ohmic internal resistance was significantly better than that of other polarization parameters, and the sensitivity was higher.

[0047] 2. Based on the extended Kalman filter algorithm, a parameter estimator is designed that identifies only two high-sensitivity parameters: open-circuit voltage and ohmic internal resistance. Other polarization parameters are identified offline using an adaptive particle swarm optimization algorithm, and the mapping relationship between each polarization parameter and state of charge (SOC) is calibrated. Open-circuit voltage U ocv The mapping relationship between SOC and OCV is determined by the OCV-SOC fitting curve. Taking a 5th-order polynomial fitting as an example, the following relationship is satisfied:

[0048]

[0049] The parameter to be identified, θ, can be expressed as R0 and U. ocv Combination of fitting parameters:

[0050] θ = [R0, m0, m1, m2, m3, m4, m5] T (2)

[0051] 3. Design a parameter estimator with an adaptive module, and combine it with the previous parameter estimator to establish the ADEKF algorithm to jointly estimate the model parameters and SOC. The parameter estimator uses the ohmic internal resistance R0 as the state variable, and the state estimator uses x... k =[U 1,k U 2,k SOC k ] T As state variables, both are represented by the terminal voltage U. L For the observables, the state-space equation of the ADEKF algorithm is:

[0052]

[0053] Where θ k x k r is a state variable k w k For process noise, v k e k For observation noise, process noise ω k r k and observation noise υ k e k All are Gaussian white noise with a mean of 0 and are mutually independent, with covariance matrices Q and Q', respectively. x Q θ R x and R θ .

[0054] The SOC estimation process based on the ADEKF algorithm is as follows:

[0055] 1) First, initialize the parameter estimator and state estimator: including x0, θ0、 and

[0056] 2) Prior estimation and prediction of secondary parameters and states, including:

[0057] State estimator time update:

[0058]

[0059] Time updates for parameter estimators:

[0060]

[0061] 3) Then, the posterior estimates and corrections for the parameters and states include:

[0062] State estimator measurement update:

[0063]

[0064] Parameter estimator measurement update:

[0065]

[0066] 4) Update of process noise and observation noise covariance of the state estimator:

[0067]

[0068] in:

[0069]

[0070] In equations (4) to (9), those marked with "-" represent the prior estimate at the current time, and those marked with "+" represent the posterior estimate. and These represent the Kalman gains of the state estimator and the parameter estimator, respectively. and Then, these represent the noise covariance matrices of the two estimators, respectively.

[0071] To verify the accuracy of the SOC estimation in this invention, the ADEKF algorithm proposed in this invention is compared with traditional algorithms based on the FUDS operating condition. The SOC estimation results and error curves are shown below. Figure 3a and Figure 3b As shown in Table 1, the advantages and disadvantages of several algorithms are compared in terms of SOC maximum error, average error, and root mean square error.

[0072] Table 1 Comparative Analysis of SOC Estimation Errors under FUDS Conditions

[0073]

[0074] from Figure 3a and Figure 3b As can be seen from Table 1, the maximum error of the ADEKF algorithm is only 1.64%, which is a further improvement compared to other algorithms. The error curve is relatively stable except for a slight decrease at the end of the discharge. Moreover, the running time is only about half that of FFRLS-EKF, which identifies all parameters, greatly reducing the complexity and computational cost of the algorithm.

[0075] To verify the robustness of the proposed method under measurement error interference, this embodiment takes current noise as an example. By introducing observation noise with a mean of 0 and a standard deviation of 0.1A, 0.2A, 0.3A and 0.4A into the current measurement data, simulation is performed based on the established algorithm model. The magnitude of the standard deviation of the current noise reflects the strength of the noise. Figure 4a and Figure 4b The SOC estimation results and error curves of the ADEKF algorithm under different current noise interferences are shown. As can be seen from the figures, when the standard deviation of the current noise is 0.1A, the relative error of SOC remains stable throughout the entire discharge cycle, fluctuating only at the end of the discharge, reaching a maximum error of -1%. As the standard deviation of the noise signal gradually increases from 0.1A to 0.4A, the overall trend of the SOC estimation curve still follows the change in the true SOC value, but shifts slightly downward, indicating that the presence of noise leads to a decrease in SOC estimation accuracy. The relative error curve shows some fluctuations at the initial moment, especially at 0.4A, but quickly stabilizes. This is the adaptive update stage of the ADEKF algorithm for the noise covariance, showing that the algorithm has good adaptability to current noise. Furthermore, it was found that when the noise continues to increase, the simulation results become unstable. At this point, the excessive current noise overwhelms the dynamic characteristics of the battery reflected in the data, causing the algorithm to fail, which is consistent with practical engineering experience. The above results are sufficient to verify that the method provided by this invention has good robustness under current noise interference.

[0076] Although embodiments of the present invention have been shown and described above, it is understood that these embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and alterations to the above embodiments within the scope of the present invention without departing from its principles and spirit. The scope of protection of the present invention is defined by the claims and their equivalents.

Claims

1. A high-sensitivity parameter and SOC combined estimation method for lithium batteries, characterized in that, First, measured data were obtained through dynamic operating condition experiments of lithium batteries. Based on the structural characteristics of the equivalent circuit model and the actual identifiability of the model parameters, high-sensitivity parameters were selected. Then, a parameter estimator that only identifies high-sensitivity parameters is designed, while the other battery model parameters are obtained using an offline parameter identification method to determine the mapping relationship between each parameter and SOC. Finally, a state estimator with an adaptive module is designed, and the parameter estimator and the state estimator are combined to establish an adaptive double extended Kalman filter algorithm to achieve joint estimation of model parameters and SOC. Specifically, the following steps are included: S1 builds an experimental platform and obtains measured data through dynamic working condition experiments of lithium batteries. Based on the structural characteristics of the equivalent circuit model, it analyzes and compares the actual identifiability of each parameter of the model and determines that the open circuit voltage and ohmic internal resistance are high-sensitivity parameters. The S2 design uses a parameter estimator that only identifies open-circuit voltage and ohmic internal resistance, and determines the open-circuit voltage through an OCV-SOC fitting curve. U ocv The mapping relationship between SOC and SOC is fitted with a 5th-order polynomial: ， Let the parameter to be identified, θ, be expressed as... R 0 and U ocv Combination of fitting parameters: ， The S3 design incorporates a state estimator with an adaptive module, which, combined with a parameter estimator, establishes the ADEKF algorithm to jointly estimate model parameters and state of charge (SOC). The adaptive module employs a Sage-Husa filter, the parameter estimator uses ohmic internal resistance as the state variable, and the state estimator uses SOC and polarization voltage as state variables. Both use terminal voltage as the observation. The state-space equation of the ADEKF algorithm is as follows: ， in θ k , x k For state variables, r k , ω k For process noise, v k , e k For observation noise, process noise ω k , r k and observation noise v k , e k All are Gaussian white noise with a mean of 0 and are mutually independent. Their covariance matrices are respectively Q x , Q θ , R x and R θ .

2. The method for jointly estimating high-sensitivity parameters and SOC of a lithium battery according to claim 1, characterized in that, The offline parameter identification methods include particle swarm optimization, least squares method, and curve fitting method.

3. The method for jointly estimating high-sensitivity parameters and SOC of a lithium battery according to claim 1, characterized in that, The SOC estimation process based on the ADEKF algorithm in step S3 is as follows: 1) Initialization of parameter estimator and state estimator: including x 0、 P 0 x , Q 0 x , R 0 x , θ 0、 P 0 θ , Q 0 θ and R 0 θ ; 2) Prior estimation and prediction of parameters and states, including: State estimator time update: ， Time updates for parameter estimators: ， 3) Posterior estimation and correction of parameters and states, including: State estimator measurement update: ， Parameter estimator measurement update: ， 4) Update of process noise and observation noise covariance of the state estimator: ， in: ， In equations (4) to (9), those marked with "-" represent the prior estimate at the current time, and those marked with "+" represent the posterior estimate. K k x and K k θ These represent the Kalman gains of the state estimator and the parameter estimator, respectively. P k x and P k θ Then, these represent the noise covariance matrices of the two estimators, respectively.

Images