In the field of electric vehicles and energy storage systems, accurate battery modeling is crucial for effective battery management, particularly for estimating the State of Charge (SOC). Among various battery types, the LiFePO4 battery has gained prominence due to its safety, longevity, and thermal stability. However, the internal electrochemical processes of a LiFePO4 battery are nonlinear and complex, making direct measurement of key parameters challenging. Therefore, developing precise equivalent circuit models is essential to simulate battery behavior and indirectly estimate internal states. Traditional models, such as the Thevenin model, often use fixed parameters and a single RC network to represent polarization effects, but they fail to capture the dynamic changes in internal parameters during charging and discharging cycles. This limitation reduces their accuracy in real-world applications. To address this, we propose a second-order variable RC equivalent circuit model that incorporates dual polarization effects and dynamically varying parameters, specifically tailored for the LiFePO4 battery. This model aims to balance complexity and precision, providing a more reliable foundation for SOC estimation and battery management systems.
The internal characteristics of a LiFePO4 battery involve both electrochemical polarization and concentration polarization, which evolve with SOC. By extending the traditional Thevenin model to include two RC parallel networks, we can better represent these dual polarization phenomena. Moreover, by allowing parameters such as ohmic resistance and polarization resistances and capacitances to vary with SOC, the model more accurately reflects the battery’s dynamic behavior. In this research, we detail the development of this second-order variable RC model for the LiFePO4 battery, including experimental procedures for parameter identification, mathematical formulations, and simulation validation. The goal is to demonstrate that this model offers superior accuracy compared to fixed-parameter models, thereby enhancing the reliability of SOC estimation for LiFePO4 batteries in practical scenarios.
To begin, let’s establish the mathematical foundation of the second-order variable RC model for the LiFePO4 battery. The equivalent circuit consists of an open-circuit voltage source \( U_{ocv} \), an ohmic resistance \( R_0 \), and two RC parallel networks representing electrochemical polarization (with resistance \( R_{p1} \) and capacitance \( C_{p1} \)) and concentration polarization (with resistance \( R_{p2} \) and capacitance \( C_{p2} \)). Unlike traditional models, all parameters except the current are functions of SOC, denoted as \( R_0(SOC) \), \( R_{p1}(SOC) \), \( R_{p2}(SOC) \), \( C_{p1}(SOC) \), and \( C_{p2}(SOC) \). The state variables are the voltages across the RC networks, \( U_{p1} \) and \( U_{p2} \), and the input is the load current \( I_L \) (positive for charging, negative for discharging). The output is the terminal voltage \( U_L \). Applying Kirchhoff’s laws, the state-space equations are derived as follows:
The state equation is given by:
$$
\begin{bmatrix} \dot{U}_{p1} \\ \dot{U}_{p2} \end{bmatrix} = \begin{bmatrix} -\frac{1}{C_{p1}(SOC) R_{p1}(SOC)} & 0 \\ 0 & -\frac{1}{C_{p2}(SOC) R_{p2}(SOC)} \end{bmatrix} \begin{bmatrix} U_{p1} \\ U_{p2} \end{bmatrix} + \begin{bmatrix} \frac{1}{C_{p1}(SOC)} \\ \frac{1}{C_{p2}(SOC)} \end{bmatrix} I_L
$$
where \( \dot{U}_{p1} \) and \( \dot{U}_{p2} \) denote the time derivatives of the polarization voltages. The output equation for the terminal voltage is:
$$
U_L = U_{ocv}(SOC) – U_{p1} – U_{p2} – I_L R_0(SOC)
$$
Alternatively, this can be expressed as:
$$
U_L = [-1 \ -1] \begin{bmatrix} U_{p1} \\ U_{p2} \end{bmatrix} + [-R_0(SOC)] I_L + U_{ocv}(SOC)
$$
The open-circuit voltage \( U_{ocv}(SOC) \) is also a function of SOC and is determined experimentally. This formulation allows the model to adapt to changes in the LiFePO4 battery’s internal state, providing a more accurate representation than fixed-parameter models. The dynamic nature of parameters is critical because, as the LiFePO4 battery cycles, its internal resistance and capacitance values shift due to factors like aging and temperature, but in this study, we focus on SOC dependence at a constant temperature of 25°C.
To identify the parameters for the LiFePO4 battery, we conducted a series of experiments, including constant current charge-discharge tests and Hybrid Pulse Power Characterization (HPPC) tests. The test object was a commercial LiFePO4 battery with a nominal voltage of 3.2V and capacity of 20Ah. All experiments were performed in a controlled environment at 25°C using a battery testing system. The constant current test involved charging the LiFePO4 battery at 0.5C (10A) to 3.6V, then switching to constant voltage until the current dropped to 0.033C (0.66A), indicating full charge (SOC=1). After resting for 1 hour, the battery was discharged at 0.5C in 10% increments, with each step followed by a 1-hour rest period, until the cutoff voltage of 2.5V was reached. The resting voltages after charge and discharge were averaged to obtain the \( U_{ocv} \)-SOC relationship, which is fundamental for modeling the LiFePO4 battery. The data was fitted using MATLAB, resulting in a polynomial curve that describes \( U_{ocv} \) as a function of SOC.
For parameter identification, HPPC tests were conducted at ten SOC points from 1.0 to 0.1. At each point, current pulses of ±20A (1C) were applied: discharge for 10 seconds, rest for 40 seconds, and charge for 10 seconds. This process captures the transient response of the LiFePO4 battery, enabling extraction of parameters like ohmic resistance and polarization time constants. The ohmic resistance \( R_0 \) was calculated from the instantaneous voltage jump when the current pulse stopped, using the formula:
$$
R_0(SOC) = \frac{|U_4 – U_3|}{I}
$$
where \( U_3 \) and \( U_4 \) are voltages before and after the current interruption, and \( I \) is the pulse current magnitude (20A). For example, at SOC=0.8, this yielded a specific value for the LiFePO4 battery. The polarization time constants \( \tau_1 \) and \( \tau_2 \) were obtained by fitting the rest period voltage recovery to a double exponential decay, representing the zero-input response of the RC networks:
$$
U_L(t) = U_{ocv} – U_{p1}(0) \exp\left(-\frac{t}{\tau_1}\right) – U_{p2}(0) \exp\left(-\frac{t}{\tau_2}\right)
$$
Using MATLAB’s curve fitting tool (cftool), we fitted this equation to the experimental data from the LiFePO4 battery, achieving high accuracy (R-square > 0.99). Similarly, the polarization resistances \( R_{p1} \) and \( R_{p2} \) were identified from the zero-state response during the current pulse phase:
$$
U_L(t) = U_{ocv} – I_L R_0 – I_L R_{p1} \left[1 – \exp\left(-\frac{t}{\tau_1}\right)\right] – I_L R_{p2} \left[1 – \exp\left(-\frac{t}{\tau_2}\right)\right]
$$
With \( \tau_1 \) and \( \tau_2 \) known, this equation was fitted to obtain \( R_{p1} \) and \( R_{p2} \). Finally, the polarization capacitances \( C_{p1} \) and \( C_{p2} \) were computed using the relation \( \tau = R C \), i.e., \( C_{p1} = \tau_1 / R_{p1} \) and \( C_{p2} = \tau_2 / R_{p2} \). This process was repeated for each SOC point, resulting in a comprehensive dataset for the LiFePO4 battery.
The identified parameters for the LiFePO4 battery at different SOC levels are summarized in Table 1. These values highlight the dynamic nature of the battery’s internal characteristics. For instance, as SOC decreases, the ohmic resistance generally increases, with a sharp rise at low SOC, which is consistent with the behavior of LiFePO4 batteries. The polarization resistances and capacitances also show significant variations, emphasizing the need for variable parameters in the model.
| SOC | \( R_0 \) (mΩ) | \( R_{p1} \) (mΩ) | \( R_{p2} \) (mΩ) | \( C_{p1} \) (F) | \( C_{p2} \) (F) |
|---|---|---|---|---|---|
| 1.0 | 3.457 | 1.745 | 3.623 | 1464 | 9630 |
| 0.9 | 3.498 | 1.796 | 4.166 | 1454 | 8548 |
| 0.8 | 3.513 | 2.096 | 3.971 | 1407 | 9955 |
| 0.7 | 3.623 | 2.037 | 3.274 | 1361 | 10714 |
| 0.6 | 3.623 | 2.232 | 3.462 | 1260 | 9929 |
| 0.5 | 3.624 | 2.420 | 3.648 | 1239 | 9520 |
| 0.4 | 3.635 | 2.432 | 5.031 | 1210 | 7588 |
| 0.3 | 3.657 | 2.515 | 6.836 | 1235 | 5908 |
| 0.2 | 3.676 | 3.039 | 7.482 | 1097 | 5663 |
| 0.1 | 4.013 | 3.327 | 9.751 | 973 | 2905 |
To incorporate these dynamics into the model, we used polynomial fitting in MATLAB to express each parameter as a function of SOC. For example, the open-circuit voltage \( U_{ocv}(SOC) \) was fitted to a 5th-order polynomial, ensuring high precision (fitting error negligible). Similarly, \( R_0(SOC) \), \( R_{p1}(SOC) \), \( R_{p2}(SOC) \), \( C_{p1}(SOC) \), and \( C_{p2}(SOC) \) were represented as polynomial functions. This allows the model to continuously adjust parameters based on the real-time SOC of the LiFePO4 battery, capturing its internal state more accurately. The polynomial equations are not listed here for brevity, but they can be derived from the data in Table 1 using standard curve-fitting techniques. This approach is particularly effective for LiFePO4 batteries due to their stable voltage plateau and predictable parameter trends.
With the parameter functions established, we built a simulation model in MATLAB/Simulink to validate the second-order variable RC model for the LiFePO4 battery. The Simulink block diagram implemented the state-space equations, with parameter blocks configured to update based on SOC input. The SOC was estimated using a coulomb counting method integrated into the simulation, though for validation, we used known SOC values from experiments. Two test scenarios were simulated: constant current discharge at 0.5C (10A) and HPPC cycles at 1C (20A). The simulation outputs for terminal voltage were compared against experimental data collected from the LiFePO4 battery during actual tests. For comparison, we also simulated a traditional fixed-parameter second-order RC model, where parameters were set to average values from Table 1.
The results demonstrated the superiority of the variable-parameter model for the LiFePO4 battery. In the constant current discharge scenario, the variable-parameter model had an average terminal voltage error of 12.36 mV, while the fixed-parameter model had an error of 31.50 mV. This represents a 60.76% improvement in accuracy for the LiFePO4 battery. In the HPPC scenario, the errors were 8.54 mV and 20.29 mV for the variable and fixed models, respectively, indicating a 57.95% accuracy enhancement. These errors are within the 10% tolerance specified in industry standards, confirming the model’s practicality for LiFePO4 battery applications. The voltage comparison curves (not shown here, but describable) revealed that the variable-parameter model closely tracked the experimental data, especially during transient phases, whereas the fixed-parameter model exhibited larger deviations, particularly at low SOC levels where parameter variations are most pronounced in LiFePO4 batteries.
To further illustrate the mathematical aspects, let’s delve into the parameter fitting process. For the LiFePO4 battery, the open-circuit voltage as a function of SOC can be expressed as a polynomial:
$$
U_{ocv}(SOC) = a_0 + a_1 SOC + a_2 SOC^2 + a_3 SOC^3 + a_4 SOC^4 + a_5 SOC^5
$$
where coefficients \( a_0 \) to \( a_5 \) are determined from experimental data. Similarly, for ohmic resistance, we have:
$$
R_0(SOC) = b_0 + b_1 SOC + b_2 SOC^2 + b_3 SOC^3 + b_4 SOC^4 + b_5 SOC^5
$$
and analogous polynomials for \( R_{p1} \), \( R_{p2} \), \( C_{p1} \), and \( C_{p2} \). These functions enable real-time adjustment in battery management systems for LiFePO4 batteries. The use of higher-order polynomials ensures fidelity, but for implementation, lower-order approximations can be used to reduce computational load, depending on the application requirements for the LiFePO4 battery.
The experimental setup for the LiFePO4 battery involved meticulous calibration to ensure data reliability. The battery testing system recorded voltage and current at high sampling rates, allowing precise extraction of transient responses. During HPPC tests, the LiFePO4 battery’s voltage curves exhibited characteristic jumps and recoveries, which were analyzed to derive time constants. For example, at SOC=0.8, the zero-input response fitting yielded \( \tau_1 = 2.945 \, \text{s} \) and \( \tau_2 = 39.82 \, \text{s} \) for the LiFePO4 battery, corresponding to the fast electrochemical polarization and slow concentration polarization, respectively. These values align with typical behaviors of LiFePO4 batteries, where diffusion processes are slower than charge transfer reactions. The parameter identification method is robust and can be applied to other LiFePO4 battery variants, though specific values may vary with chemistry and design.
In terms of simulation, the Simulink model for the LiFePO4 battery incorporated several key blocks: a controlled current source for load profiles, integrators for state variables, and function blocks for parameter updates. The state equations were solved using ode4 (Runge-Kutta) solver with a fixed step size of 0.1 seconds to balance accuracy and speed. The simulation output was validated against experimental data points, and error metrics were computed using root mean square error (RMSE) and mean absolute error (MAE). For the LiFePO4 battery, the variable-parameter model achieved RMSE values below 15 mV across all tests, meeting the precision criteria for SOC estimation algorithms. This makes it suitable for integration into battery management systems for electric vehicles using LiFePO4 batteries.
Beyond simulation, the variable-parameter model for the LiFePO4 battery has implications for adaptive control and prognostics. By accurately capturing internal dynamics, it can enhance SOC estimation techniques such as Kalman filtering or neural networks. For instance, in an extended Kalman filter (EKF) setup, the model’s state equations can be linearized around the operating point, with parameter functions providing Jacobian matrices. This improves convergence and reduces SOC estimation errors for LiFePO4 batteries under varying load conditions. Additionally, the model can be extended to include temperature effects, though this study focused on isothermal conditions at 25°C. Future work could incorporate thermal dependencies to further refine the model for LiFePO4 batteries in extreme environments.
The advantages of the second-order variable RC model for LiFePO4 batteries are manifold. First, it maintains a balance between complexity and accuracy—adding more RC networks might marginally improve precision but at the cost of increased computational burden, as noted in literature. Second, the variable parameters inherently account for aging effects to some extent, as SOC-dependent trends may shift over the battery’s lifecycle. For LiFePO4 batteries, which are known for their long cycle life, this adaptability is valuable. Third, the model is grounded in experimental data from standard tests like HPPC, making it practical for industry adoption. Companies manufacturing or using LiFePO4 batteries can implement this model with relative ease, using similar testing protocols.
To summarize the methodology, we outline the steps for developing the variable-parameter model for a LiFePO4 battery:
- Perform constant current charge-discharge tests to obtain the \( U_{ocv} \)-SOC relationship.
- Conduct HPPC tests at multiple SOC points to capture transient responses.
- Identify parameters \( R_0 \), \( R_{p1} \), \( R_{p2} \), \( C_{p1} \), and \( C_{p2} \) at each SOC using curve fitting.
- Fit polynomial functions to describe each parameter as a function of SOC.
- Build a Simulink model implementing the state-space equations with variable parameters.
- Validate the model against experimental data under various load profiles.
This process ensures a data-driven approach tailored for LiFePO4 batteries. The model’s efficacy is demonstrated through error reduction, and it can be a cornerstone for advanced battery management systems.
In conclusion, the second-order variable RC equivalent circuit model proposed for LiFePO4 batteries offers significant improvements over traditional fixed-parameter models. By incorporating dual polarization effects and SOC-dependent parameter variations, it accurately replicates the static and dynamic behaviors of LiFePO4 batteries. Experimental validation using HPPC tests and simulation in Simulink confirms its precision, with voltage errors reduced by over 57% compared to conventional models. This model strikes an optimal balance between simplicity and accuracy, making it suitable for real-time applications in electric vehicles and energy storage systems utilizing LiFePO4 batteries. Future research could explore integration with machine learning for parameter adaptation or extension to battery pack modeling. Ultimately, this work contributes to the reliable management of LiFePO4 batteries, enhancing their performance and longevity in sustainable energy solutions.
For reference, the key formulas and tables are encapsulated above. The polynomial coefficients for parameter functions can be stored in lookup tables in embedded systems for efficient computation. The LiFePO4 battery’s characteristics, such as flat voltage discharge curves and low internal resistance, are well-captured by this model, underscoring its relevance. As the demand for LiFePO4 batteries grows in applications ranging from portable electronics to grid storage, accurate modeling tools like this will play a pivotal role in optimizing their operation and maintenance.