The precise and reliable estimation of the State of Charge (SOC) is universally recognized as a cornerstone function within any advanced Battery Management System (BMS). For electric vehicles and a multitude of other applications powered by lithium-ion batteries, an accurate SOC reading is not merely a convenience but a critical necessity. It safeguards the battery against harmful overcharge and deep discharge conditions, directly informs range predictions, and is fundamental to optimizing performance and longevity. Given that the SOC is an internal, non-measurable state, its value must be inferred indirectly through mathematical models and estimation algorithms based on measurable quantities like terminal voltage, current, and temperature. Consequently, the fidelity of the underlying battery model and the robustness of the estimation algorithm become paramount determinants of overall system performance and safety. My research focuses on advancing this field by addressing key limitations in traditional modeling and estimation approaches.
Conventional methods for modeling the dynamic behavior of a lithium-ion battery predominantly rely on integer-order Equivalent Circuit Models (ECMs). These models typically comprise networks of ideal resistors and capacitors, such as the Thevenin model or the dual-polarization (2RC) model. While computationally efficient, these integer-order frameworks possess inherent shortcomings. They often struggle to accurately capture the complex, frequency-dependent electrochemical phenomena occurring within the cell, such as the distribution of relaxation times associated with charge transfer double-layer effects and solid-state diffusion processes. The assumption of ideal, integer-order capacitors can lead to discrepancies between simulated and actual voltage responses, particularly under dynamic load profiles.
To overcome these limitations, I propose a paradigm shift towards fractional-order calculus for lithium-ion battery modeling. Fractional-order models offer a more nuanced and physically representative toolset. By replacing ideal capacitors with Constant Phase Elements (CPEs), characterized by a fractional-order impedance, the model gains the ability to more faithfully represent the depressed semicircles and Warburg diffusion tails observed in real electrochemical impedance spectroscopy (EIS) data. This leads to a superior fit of the battery’s transient voltage behavior. The foundation of this approach lies in the definitions of fractional calculus. For a continuous function \( f(t) \), the Grünwald–Letnikov (G-L) definition of the fractional derivative of order \( \alpha \) is particularly suitable for digital implementation:
$$ _{a}^{GL}D_{t}^{\alpha} f(t) = \lim_{h \to 0} h^{-\alpha} \sum_{j=0}^{\left[ \frac{t-a}{h} \right]} (-1)^j \binom{\alpha}{j} f(t – jh) $$
where \( \alpha \) is the fractional order, \( a \) is the initial time, and \( h \) is the step size. This definition forms the mathematical bedrock for describing the dynamics of circuit elements with non-integer order.
The specific fractional-order equivalent circuit model adopted in my work is depicted conceptually below. It consists of an ohmic resistor \( R_0 \), and two parallel \( R-CPE \) branches in series. The first branch (\( R_1 \parallel CPE_1 \)) models the charge transfer and double-layer polarization effects, while the second branch (\( R_2 \parallel CPE_2 \)) captures the diffusion dynamics within the lithium-ion battery. The CPE’s impedance is given by \( Z_{CPE} = 1 / [C (j\omega)^{\alpha}] \), where \( C \) is the pseudo-capacitance and \( \alpha \) is the fractional exponent (\( 0 < \alpha \leq 1 \)).
The terminal voltage \( U_t(t) \) of this fractional-order model for a lithium-ion battery is governed by the following equations, derived from Kirchhoff’s laws:
$$
\begin{aligned}
I(t) &= \frac{U_{CPE_1}(t)}{Z_{CPE_1}} + \frac{U_1(t)}{R_1} = \frac{U_{CPE_2}(t)}{Z_{CPE_2}} + \frac{U_2(t)}{R_2} \\
U_{ocv}(SOC(t)) &= U_t(t) + I(t)R_0 + U_1(t) + U_2(t)
\end{aligned}
$$
where \( U_{ocv} \) is the open-circuit voltage, a known function of SOC, and \( I(t) \) is the applied current. The fractional-order differential equations for the polarization voltages are:
$$ D^{\alpha_1} U_1(t) = -\frac{U_1(t)}{R_1 C_1} + \frac{I(t)}{C_1}, \quad D^{\alpha_2} U_2(t) = -\frac{U_2(t)}{R_2 C_2} + \frac{I(t)}{C_2} $$
Accurate model parameters are prerequisite for high-fidelity SOC estimation. The parameter set \( \theta = [R_0, R_1, C_1, \alpha_1, R_2, C_2, \alpha_2] \) must be identified from experimental data. I employ an Adaptive Genetic Algorithm (AGA) for this high-precision offline identification. The AGA optimizes the parameters by minimizing the error between the model output voltage \( U_T \) and the measured voltage \( U_R \) over a dataset of length \( M \). The fitness function \( J \) for the genetic algorithm is the Sum of Squared Errors (SSE):
$$ J(\theta) = \sum_{j=1}^{M} \left[ U_R(j) – U_T(j, \theta) \right]^2 $$
The adaptive mechanism dynamically adjusts the crossover and mutation probabilities based on population fitness, preventing premature convergence to local minima and ensuring a thorough exploration of the parameter space for the lithium-ion battery model. The performance of the integer-order (2RC) model and the fractional-order model, both parameterized via AGA, can be compared using key metrics as summarized below:
| Model Type | Parameter Set | SSE (Typical Value) | Remarks |
|---|---|---|---|
| Integer-Order (2RC) | \( [R_0, R_1, C_1, R_2, C_2] \) | Higher | Assumes ideal capacitors (\(\alpha=1\)). Struggles with dispersive dynamics. |
| Fractional-Order | \( [R_0, R_1, C_1, \alpha_1, R_2, C_2, \alpha_2] \) | Significantly Lower | CPE provides better fit to EIS. AGA effectively identifies fractional orders \(\alpha_i\). |
With a high-fidelity fractional-order model in place, the next challenge is real-time SOC estimation. The SOC is defined as the ratio of remaining charge to nominal capacity:
$$ SOC(t) = SOC(t_0) – \frac{1}{Q_n} \int_{t_0}^{t} \eta I(\tau) d\tau $$
where \( Q_n \) is the nominal capacity and \( \eta \) is the coulombic efficiency. Combining this with the fractional-order model state equations leads to a discrete-time state-space representation suitable for filtering. The state vector is chosen as \( \mathbf{x}_k = [SOC_k, U_{1,k}, U_{2,k}]^T \). The nonlinear state update and measurement equations are:
$$
\begin{aligned}
\mathbf{x}_{k+1} &= f(\mathbf{x}_k, I_k) + \mathbf{w}_k \\
U_{t,k} &= U_{ocv}(SOC_k) – U_{1,k} – U_{2,k} – I_k R_0 + v_k
\end{aligned}
$$
where \( \mathbf{w}_k \) and \( v_k \) are process and measurement noise, respectively, assumed to be zero-mean with covariance matrices \( \mathbf{Q}_k \) and \( R_k \). The nonlinearity stems from the \( U_{ocv}(SOC) \) relationship.
The standard Fractional-Order Extended Kalman Filter (FOEKF) linearizes these equations around the current state estimate. However, its performance is highly sensitive to the pre-defined noise covariance matrices \( \mathbf{Q} \) and \( R \). Inaccurate initial guesses for these statistics can lead to slow convergence, large estimation errors, or even filter divergence—a significant risk in practical BMS applications for lithium-ion batteries where operating conditions and noise characteristics are variable.
To robustify the estimator, I have designed and implemented a Fractional-Order Adaptive Extended Kalman Filter (FOAEKF). This algorithm integrates the fractional-order battery dynamics with an adaptive covariance matching technique. The core idea is to use the innovation sequence—the difference between the actual measurement and its prediction—to continuously estimate and update the noise statistics online. The sample covariance of the innovation sequence over a sliding window of size \( N \) is computed:
$$ \hat{\mathbf{C}}_k = \frac{1}{N} \sum_{i=k-N+1}^{k} \boldsymbol{\gamma}_i \boldsymbol{\gamma}_i^T $$
where \( \boldsymbol{\gamma}_k = U_{t,k} – \hat{U}_{t,k|k-1} \) is the innovation. By equating this empirical covariance to its theoretical expression \( \mathbf{H}_k \mathbf{P}_{k|k-1} \mathbf{H}_k^T + \hat{R}_k \), we can derive an update law for the measurement noise covariance estimate \( \hat{R}_k \). A similar logic can be applied to adapt \( \mathbf{Q}_k \). This feedback mechanism allows the FOAEKF to self-tune in response to changing noise environments, enhancing the robustness and accuracy of SOC estimation for the lithium-ion battery. The step-by-step procedure for the FOAEKF is outlined below:
Step 1: Initialization
Set initial state estimate \(\hat{\mathbf{x}}_{0|0}\) and error covariance \(\mathbf{P}_{0|0}\).
Step 2: State & Covariance Prediction (Time Update)
$$ \hat{\mathbf{x}}_{k|k-1} = f(\hat{\mathbf{x}}_{k-1|k-1}, I_{k-1}) $$
$$ \mathbf{P}_{k|k-1} = \mathbf{A}_{k-1} \mathbf{P}_{k-1|k-1} \mathbf{A}_{k-1}^T + \mathbf{Q}_{k-1} $$
where \(\mathbf{A}_{k-1}\) is the Jacobian of \(f\) w.r.t. the state.
Step 3: Noise Covariance Adaptation
Compute innovation \(\boldsymbol{\gamma}_k\). Update estimates for \(\hat{R}_k\) and potentially \(\hat{\mathbf{Q}}_k\) using the covariance matching principle.
Step 4: Kalman Gain Calculation
$$ \mathbf{K}_k = \mathbf{P}_{k|k-1} \mathbf{H}_k^T (\mathbf{H}_k \mathbf{P}_{k|k-1} \mathbf{H}_k^T + \hat{R}_k)^{-1} $$
where \(\mathbf{H}_k\) is the Jacobian of the measurement equation w.r.t. the state.
Step 5: State & Covariance Update (Measurement Update)
$$ \hat{\mathbf{x}}_{k|k} = \hat{\mathbf{x}}_{k|k-1} + \mathbf{K}_k \boldsymbol{\gamma}_k $$
$$ \mathbf{P}_{k|k} = (\mathbf{I} – \mathbf{K}_k \mathbf{H}_k) \mathbf{P}_{k|k-1} $$
The performance of the proposed methodology was rigorously validated using experimental data from a ternary lithium-ion battery subjected to the Urban Dynamometer Driving Schedule (UDDS) profile, a highly dynamic discharge sequence. The following table presents a quantitative comparison of the estimation errors for different model-algorithm combinations, clearly demonstrating the advantages of the fractional-order adaptive approach.
| Test Case | Algorithm | Model | RMSE (SOC) | Maximum Absolute Error | Key Observation |
|---|---|---|---|---|---|
| UDDS (Exact Noise Stats) | IO-EKF | Integer-Order (2RC) | 0.0148 | 0.0351 | Baseline performance with integer-order dynamics. |
| FOEKF | Fractional-Order | 0.0112 | 0.0275 | Fractional model provides lower error and faster initial convergence. | |
| UDDS (Mismatched Noise Stats) | FOEKF | Fractional-Order | 0.0126 | 0.0310 | Performance degrades with incorrect Q, R. |
| FOAEKF | Fractional-Order | 0.0108 | 0.0258 | Adaptive mechanism recovers accuracy, achieving the best overall performance. |
The results are unequivocal. The fractional-order model consistently outperforms its integer-order counterpart when used with the same filtering algorithm (EKF), as evidenced by the lower RMSE. This confirms that the fractional-order calculus more accurately captures the inherent dynamics of the lithium-ion battery. More importantly, the FOAEKF algorithm demonstrates superior robustness. In the scenario with deliberately mismatched initial noise covariances, the standard FOEKF’s error increases. In contrast, the FOAEKF successfully adapts to the correct noise levels, ultimately achieving the lowest RMSE of all tested configurations. This adaptive capability is crucial for real-world deployment where sensor noise characteristics and model uncertainties are not perfectly known a priori.
In conclusion, accurate state-of-charge estimation remains a vital and challenging task for ensuring the safety, efficiency, and longevity of systems powered by lithium-ion batteries. This research has systematically addressed key limitations in the prevailing SOC estimation pipeline. By adopting a fractional-order equivalent circuit model, I have achieved a more physically representative characterization of the battery’s electrochemical dynamics compared to traditional integer-order models. The use of an Adaptive Genetic Algorithm enables high-precision offline identification of the model parameters, including the fractional orders. To tackle the critical issue of uncertainty in noise statistics, I have designed and implemented a Fractional-Order Adaptive Extended Kalman Filter. This algorithm synergistically combines the fidelity of fractional-order modeling with online noise covariance estimation, resulting in an estimator that is both accurate and robust. Comprehensive validation under demanding UDDS driving cycles confirms that the proposed FOAEKF framework delivers superior convergence speed and estimation accuracy for the lithium-ion battery SOC, marking a significant step forward towards more reliable and intelligent battery management systems. Future work will focus on extending this framework to include joint state and parameter estimation for adapting to battery aging, as well as implementation and testing on embedded BMS hardware.