The ambitious “dual-carbon” targets have unequivocally set the strategic course for China’s energy transition, creating unprecedented opportunities for integrating renewable energy sources. However, this paradigm shift towards a new power system, dominated by renewables and power electronics, introduces significant challenges for maintaining grid stability. The inherent intermittency and volatility of renewable generation lead to substantial frequency fluctuations. Concurrently, the declining system inertia, as synchronous generators are displaced, exacerbates the difficulty of achieving real-time power balance. Traditional frequency regulation resources, such as thermal power plants, with their slow mechanical response times often exceeding one minute, are increasingly inadequate to meet the stringent speed and accuracy demands of modern grid control.
Electrochemical energy storage, particularly systems based on energy storage cell technology, has emerged as a transformative solution. Lithium-ion battery energy storage cell systems offer rapid, precise, and bi-directional power control, capable of responding to Automatic Generation Control (AGC) signals within one second. This performance is orders of magnitude faster than conventional units, making even a relatively small capacity of battery storage highly effective in enhancing the overall frequency regulation capability of a power system. Consequently, the scientific modeling and control of lithium-ion energy storage cell systems, especially when co-located with renewable generation for frequency regulation services, have become a critical research focus. This article delves into the precise modeling, parameter identification, and state estimation of lithium-ion energy storage cells, providing a foundation for their optimal deployment in grid frequency support.
1. Advanced Modeling of the Lithium-Ion Energy Storage Cell
The cornerstone of effective system control is an accurate mathematical model that describes the relationship between a battery’s operational characteristics and influencing factors such as State of Charge (SOC), terminal voltage, current, and temperature. Among various modeling approaches, equivalent circuit models (ECMs) are widely favored for their strong descriptive capability, adaptability, and relative simplicity for control integration. To comprehensively capture the DC and dynamic AC characteristics of a lithium-ion energy storage cell, an improved composite equivalent circuit model is proposed. This model intricately links the cell’s capacity dynamics with its electrical transient response.
The left sub-circuit in the model tracks the cell’s state. The capacitor \( C_{capacity} \) represents the usable capacity of the energy storage cell, which is a variable affected by aging and operational conditions. The resistor \( R_{self-discharge} \) models the inherent leakage, and the voltage source \( V_{SOC} \) is the output linked to the SOC. The core equation governing SOC calculation via the Coulomb-counting method is:
$$SOC = SOC_{init} – \frac{1}{C_{use}} \int I_{batt} dt$$
where \( SOC_{init} \) is the initial state, \( I_{batt} \) is the terminal current, and \( C_{use} \) is the real-time usable capacity. The right sub-circuit describes the dynamic electrical response. The voltage source \( V_{oc}(SOC) \) is the open-circuit voltage, a primary function of SOC. The resistor \( R_{series} \) represents the lumped ohmic resistance, and \( R_{cyc} \) accounts for the increase in internal resistance due to cycling aging. The two RC parallel networks model the short-term and long-term transient voltage responses, corresponding to charge transfer and diffusion polarization phenomena, respectively. The two sides of the model are coupled through the SOC. The state-space representation of the electrical circuit is given by:
$$
\begin{aligned}
\dot{U}_{tran-s} &= -\frac{U_{tran-s}}{R_{tran-s}C_{tran-s}} + \frac{I_{batt}}{C_{tran-s}} \\
\dot{U}_{tran-l} &= -\frac{U_{tran-l}}{R_{tran-l}C_{tran-l}} + \frac{I_{batt}}{C_{tran-l}} \\
V_{batt} &= V_{oc}(SOC) – U_{tran-s} – U_{tran-l} – I_{batt}(R_{series} + R_{cyc})
\end{aligned}
$$
where \( U_{tran-s} \) and \( U_{tran-l} \) are the voltages across the two RC networks.
2. Parameter Identification and Determination for the Energy Storage Cell Model
2.1 Determining Usable Capacity and SOC
The accuracy of the Coulomb-counting method hinges on the correct value of \( C_{use} \). This parameter is not constant; it decays over time due to cycle aging and calendar aging and is instantly influenced by operating temperature and current rate (C-rate).
Cycle Aging: Capacity fade from repeated charging/discharging is often irreversible and approximately linear with cycle count. It is accelerated at higher temperatures, a relationship well-described by the Arrhenius equation. An empirical model can be formulated as:
$$C_{cyc\%} = A \cdot \exp\left(-\frac{E_a}{R \cdot T}\right) \cdot (I_{PR})^z$$
where \( C_{cyc\%} \) is the capacity loss ratio, \( A \) is a constant, \( E_a \) is activation energy, \( R \) is the gas constant, \( T \) is temperature in Kelvin, \( I_{PR} \) is the cumulative current throughput, and \( z \) is an experimental parameter.
Calendar Aging: This refers to capacity loss during storage, primarily due to self-discharge. It depends on temperature, SOC, and storage duration. Empirical piecewise functions fitted from experimental data are effective:
$$
C_{sto\%}(T) =
\begin{cases}
a_1T + b_1 \log_{10}(t) – c_1T + d_1 & T \leq 40^\circ C\\
a_2T + b_2 \log_{10}(t) – c_2T + d_2 & T > 40^\circ C
\end{cases}
$$
Temperature and C-rate Effects: Instantaneous available capacity increases with temperature but saturates above a certain point. Conversely, higher discharge C-rates reduce the usable capacity due to increased polarization, a phenomenon described by the modified Peukert’s equation:
$$C_I = C_s \left( \frac{I_s}{I} \right)^{n-1}$$
where \( C_s \) is capacity at standard current \( I_s \), and \( n \) is the battery constant.
Thus, the comprehensive usable capacity for an energy storage cell that has undergone \( n \) cycles and \( t \) days of storage, operating at temperature \( T \) and current \( I_c \), is:
$$
C_{use} = Q_0(1 – C_{cyc\%})(1 – C_{sto\%}) \cdot \frac{e^{k_1\left(\frac{1}{T_{ref}} – \frac{1}{T}\right)}}{(I_c / I_s)^{n-1}}
$$
2.2 Identification of Electrical Circuit Parameters
The parameters \( V_{oc}, R_{series}, R_{tran-s}, C_{tran-s}, R_{tran-l}, C_{tran-l}, R_{cyc} \) are identified through standardized tests like the Hybrid Pulse Power Characterization (HPPC).
Open-Circuit Voltage (OCV): The relationship \( V_{oc}(SOC) \) is foundational. It is obtained by measuring the steady-state voltage after long rest periods at various SOC points. A combined polynomial-exponential function provides a good fit:
$$V_{oc}(SOC) = a_0 + a_1SOC + a_2 \exp(a_3 SOC) + a_4 \exp\left(\frac{a_5}{1-SOC}\right)$$
Dynamic Impedance Parameters: From the voltage response to a current pulse, the ohmic resistance is found from the instantaneous voltage jump: \( R_{series} = (V_0 – V_1)/I \). The transient response curve is fitted to a double-exponential function \( V(t) = V_1 – a(1-e^{-t/\tau_s}) – b(1-e^{-t/\tau_l}) \). The RC parameters are then derived:
$$
\begin{aligned}
R_{tran-s} &= a/I, \quad C_{tran-s} = \tau_s / R_{tran-s} \\
R_{tran-l} &= b/I, \quad C_{tran-l} = \tau_l / R_{tran-l}
\end{aligned}
$$
These parameters are also functions of SOC. A common representation is:
$$
\begin{aligned}
R_{series}(SOC) &= b_0 + b_1SOC + b_2SOC^2 + b_3 \exp(b_4 SOC) + b_5 \exp(b_6 SOC)\\
R_{tran-s}(SOC) &= c_0 + c_1 e^{c_2 SOC}
\end{aligned}
$$
Influence of Temperature, C-rate, and Cycling: To enhance model fidelity, parameters are further expressed as functions of temperature \( T \) and C-rate \( I_c \). For instance, the temperature dependence can be modeled as additive corrections:
$$
R_{series}(SOC,T) = R_{series}(SOC) + \alpha_1 \exp\left(\frac{\alpha_2}{T}\right)
$$
Alternatively, for control simplicity, the composite effect of temperature and C-rate can be lumped into a single voltage correction term \( \Delta E(T, I_c) \) added to the model output. The cycle-dependent resistance increase is often modeled empirically: \( R_{cyc} = k N^z \), where \( N \) is cycle count and \( k, z \) are coefficients.
| Parameter | Primary Influence | Secondary Influence | Identification Method |
|---|---|---|---|
| \( C_{use} \) | Cycle Count, Storage Time | Temperature, C-rate | Long-term aging tests, HPPC at various T |
| \( V_{oc} \) | State of Charge (SOC) | Negligible | OCV-SOC test with long relaxation |
| \( R_{series} \) | SOC, Temperature | Cycling (via \( R_{cyc} \)) | Instantaneous voltage step in HPPC |
| \( R_{tran}, C_{tran} \) | SOC | Temperature, C-rate (minor) | Curve fitting of transient voltage in HPPC |
3. Aggregation Modeling for Energy Storage Cell Systems
Real-world grid applications require large-scale battery energy storage systems (BESS) comprising hundreds or thousands of individual energy storage cells connected in series and parallel. Modeling such a system necessitates considering cell-to-cell variations in parameters like capacity, impedance, and initial SOC, which arise from manufacturing tolerances and uneven operating conditions.
The core principle for aggregating models of individual energy storage cells is based on the laws of series and parallel circuits. A string of series-connected cells can be等效 (equivalent) to a single cell model with summed voltages and shared current, provided the SOC dynamics are carefully averaged or the weakest cell is considered for safe operation. Similarly, parallel strings can be等效 to a single cell with summed capacity (current capability) and averaged voltage.
Key factors in system-level modeling include:
- Current Imbalance in Parallel Strings: Due to slight differences in internal resistance and voltage, currents may not be evenly shared, affecting overall loss and aging.
- Voltage Imbalance in Series Strings: Capacity or impedance variation leads to different voltage drops across cells during operation, limiting the usable energy of the entire string to that of the weakest energy storage cell.
- Thermal Coupling: Cells influence each other’s temperature, creating a feedback loop on their individual parameters.
Therefore, a high-fidelity BESS model might consist of multiple instances of the single energy storage cell model, connected with balancing network models, all within a thermal network. For system-level control studies, a simplified aggregated model with parameters representing the average or worst-case behavior of the pack is often used.
| Modeling Approach | Description | Complexity | Use Case |
|---|---|---|---|
| Detailed Cell-by-Cell | Each cell modeled individually with its own parameters. | Very High | Detailed aging analysis, thermal runaway studies. |
| Average Model | One等效 cell model with averaged parameters from all cells. | Low | System-level power flow and frequency regulation studies. |
| Worst-Cell Model | Model parameters represent the weakest cell in the pack. | Low | Safe operating limit determination, conservative control. |
| Cluster Model | Cells grouped into clusters (e.g., per module) with similar states. | Medium | Balance management system design, efficiency analysis. |
4. State of Charge Estimation Under Time-Varying Parameters
Accurate real-time knowledge of the SOC is paramount for safe and efficient operation of the energy storage cell. The Coulomb-counting method is simple but drifts due to cumulative current measurement errors and an inaccurate \( C_{use} \). The model-based OCV method requires long rest periods, which are unavailable during frequency regulation duty. Therefore, advanced estimation filters are essential.
The Extended Kalman Filter (EKF) is a prominent choice for estimating the state of nonlinear systems like a battery. It combines the model prediction with real-time voltage measurements to provide an optimal estimate of the hidden states, such as SOC and the RC network voltages. The nonlinear state-space model for the energy storage cell is:
$$
\begin{aligned}
\mathbf{x}_{k+1} &= f(\mathbf{x}_k, I_k) + \mathbf{w}_k \\
y_k &= V_{batt,k} = h(\mathbf{x}_k) + v_k
\end{aligned}
$$
where \( \mathbf{x}_k = [SOC_k, U_{tran-s,k}, U_{tran-l,k}]^T \) is the state vector, \( I_k \) is the input current, \( y_k \) is the measured voltage, and \( \mathbf{w}_k, v_k \) are process and measurement noise. The functions \( f(\cdot) \) and \( h(\cdot) \) are derived from the coupled Coulomb-counting and electrical circuit equations.
The EKF linearizes these functions around the current state estimate to compute the Jacobian matrices \( \mathbf{A}_k \) and \( \mathbf{H}_k \):
$$
\mathbf{A}_k = \left. \frac{\partial f}{\partial \mathbf{x}} \right|_{\hat{\mathbf{x}}_k^-}, \quad \mathbf{H}_k = \left. \frac{\partial h}{\partial \mathbf{x}} \right|_{\hat{\mathbf{x}}_k^-}
$$
The Kalman gain \( \mathbf{K}_k \) is then calculated, which optimally weights the model prediction against the measurement innovation:
$$
\mathbf{K}_k = \mathbf{P}_k^- \mathbf{H}_k^T (\mathbf{H}_k \mathbf{P}_k^- \mathbf{H}_k^T + \mathbf{R})^{-1}
$$
Finally, the state estimate is updated:
$$
\hat{\mathbf{x}}_k^+ = \hat{\mathbf{x}}_k^- + \mathbf{K}_k \left( y_k – h(\hat{\mathbf{x}}_k^-) \right)
$$
To further improve robustness against model inaccuracies from varying C-rate and temperature, a dual estimation strategy can be employed. This involves using a normalized voltage-SOC characteristic curve that is adjusted in real-time based on the measured OCV (estimated during quiet periods) and the known temperature and C-rate effects. The normalized curve, scaled by the instantaneous \( V_{oc}(SOC) \) relationship, provides a continuously calibrated reference for the EKF, leading to highly accurate SOC estimation even under the demanding, variable-power profile of frequency regulation.
5. Simulation Analysis and Validation
The proposed modeling and estimation framework was implemented and validated in a MATLAB/Simulink environment. The model parameters were identified for a commercial 18650-type Lithium Iron Phosphate (LFP) energy storage cell.
Static Discharge Characteristics: Simulations at various constant discharge C-rates and temperatures were conducted. The model accurately reproduced the characteristic flat discharge plateau of LFP chemistry and the increased available capacity at higher temperatures, closely matching empirical data trends.
Capacity Fade Simulation: The calendar and cycle aging models were simulated over time. The simulation results for capacity loss ratio under different storage temperatures and cycle counts demonstrated a close alignment with experimental degradation data, validating the empirical aging functions.
Dynamic Pulse Response: The model was subjected to a complex dynamic current profile, typical of frequency regulation duty, featuring high-rate charge and discharge pulses. The simulated terminal voltage response was compared against measured data from a real cell under the same profile. The results showed excellent agreement, confirming the model’s ability to capture the dynamic polarization effects represented by the RC networks.
State of Charge Estimation: The performance of the EKF-based SOC estimator was tested using a realistic frequency regulation workload cycle. The cell underwent repeated charge/discharge phases at varying power levels (e.g., 10 kW discharge, 4 kW discharge, 6 kW discharge, followed by 2 kW, 10 kW, and 8 kW charge). The true SOC (calculated offline with high precision) was compared against the EKF’s estimated SOC. The estimator demonstrated high accuracy with minimal error throughout the test, effectively tracking the SOC despite the highly variable current and the model’s time-varying parameters.
The simulation results conclusively verify that the proposed comprehensive model for the lithium-ion energy storage cell provides a high-fidelity representation of its steady-state and dynamic behavior under conditions relevant to grid frequency regulation. Furthermore, the integrated EKF-based estimation strategy proves to be a robust and accurate method for real-time SOC determination, which is a critical input for any advanced battery management and grid control algorithm.
6. Conclusion
This article presents a holistic approach to modeling and simulating lithium-ion energy storage cells for grid frequency regulation applications. The core contributions are:
- An Advanced Equivalent Circuit Model: A comprehensive model was developed that integrates capacity fade mechanisms (cycle and calendar aging) and operational effects (temperature, C-rate) with a dynamic electrical network, offering a high-accuracy representation of the energy storage cell.
- A Systematic Parameter Identification Methodology: Detailed procedures were outlined for determining all model parameters from standard and characterization tests, accounting for their dependencies on SOC, temperature, and aging.
- Robust State Estimation: An Extended Kalman Filter framework, potentially augmented with adaptive normalization techniques, was proposed to accurately estimate the critical State of Charge in real-time amidst changing operational conditions.
- Simulation-Based Validation: The model and estimation methods were successfully implemented and validated through detailed simulations, confirming their accuracy against static, dynamic, and aging characteristics of a real energy storage cell.
The precise modeling and state estimation of the fundamental energy storage cell unit form the essential foundation for optimizing the control and dispatch of large-scale battery energy storage systems. This work enables more reliable and effective participation of energy storage in grid frequency services, thereby supporting the stable integration of high shares of renewable energy and the transition toward a more resilient and sustainable power system. Future work will focus on integrating this cell-level model into full system-level controllers and validating its performance in hardware-in-the-loop tests with real grid frequency signals.