The increasing integration of renewable energy sources into modern power grids, while crucial for decarbonization, presents significant challenges to system stability. A primary concern is the reduction of overall system inertia, making the grid more susceptible to frequency deviations following disturbances. The inherent variability of sources like wind and solar necessitates robust and fast-responding frequency support mechanisms. In this context, the cell energy storage system, particularly battery-based solutions, has emerged as a pivotal technology for primary frequency control (PFR). Its capability for near-instantaneous bidirectional power adjustment makes it an ideal supplement to the slower, traditional generation-based frequency response.
Traditional control strategies for engaging a cell energy storage system in PFR, such as fixed or state-of-charge (SOC)-dependent inertia and droop control, often exhibit limitations. They may not adequately suppress the rate of frequency decline (nadir) or may inadvertently hinder the recovery process. Furthermore, a critical oversight in many strategies is the assumption of perfect SOC knowledge. Inaccurate SOC estimation can lead to over-charging or over-discharging of the battery, accelerating degradation and potentially causing failure, which directly compromises the reliability of the cell energy storage system as a grid support asset. Therefore, enhancing both the adaptive capability of the control output and the accuracy of the SOC estimation is paramount for optimizing the performance and longevity of a cell energy storage system in frequency regulation services.
This article presents a comprehensive adaptive primary frequency control strategy specifically designed for battery-based cell energy storage system applications. The strategy is built upon three synergistic pillars designed to address the aforementioned shortcomings holistically, ensuring the cell energy storage system operates at its highest potential.
1. Integrated Frequency Control Strategy: Combining Trend, Inertia, and Droop
The foundational element of the proposed strategy is an enhanced control law that moves beyond conventional inertia and droop. We introduce a Trend Control component, forming a tri-modal control structure. The total power command (ΔP_E) for the cell energy storage system is given by:
$$ \Delta P_E = M_b \frac{d\Delta f}{dt} + K_b \Delta f + T_b \frac{d^2\Delta f}{dt^2} $$
Where:
– \(\Delta f\) is the grid frequency deviation (Hz).
– \(d\Delta f/dt\) is the Rate of Change of Frequency (RoCoF).
– \(d^2\Delta f/dt^2\) is the frequency acceleration.
– \(M_b\) is the adaptive inertia coefficient.
– \(K_b\) is the adaptive droop coefficient.
– \(T_b\) is the adaptive trend coefficient.
The innovation lies in the third term. The Trend Control responds directly to the acceleration of the frequency deviation. During the initial stage of a disturbance, a large negative frequency acceleration indicates a rapidly worsening situation. A positive \(T_b\) term generates additional discharge power, proactively countering the accelerating decline and effectively “flattening” the frequency dip. Conversely, during recovery, if frequency is returning too sluggishly (positive acceleration towards zero), the trend term can provide a corrective nudge. The interplay of these three signals—deviation, its first derivative (RoCoF), and its second derivative (acceleration)—provides a more complete picture of the grid’s dynamic state, allowing the cell energy storage system to respond more intelligently.
2. Full Adaptive Coefficient Scheduling
To maximize the effectiveness of the tri-modal control, static coefficients are insufficient. The coefficients \(M_b\), \(K_b\), and \(T_b\) are dynamically adapted based on real-time grid conditions and the internal state of the cell energy storage system.
2.1 Adaptive Coefficient Laws:
Each coefficient is defined as the sum of a fixed baseline and an adaptive component scaled by an SOC-dependent constraint factor (\(P_c\) for charging, \(P_d\) for discharging).
$$ M_b = \begin{cases}
M_0 + M_a, & \text{for } \Delta f < 0 \text{ (discharge)} \\
M_0 + M_a \cdot P_c, & \text{for } \Delta f > 0 \text{ (charge)}
\end{cases} $$
$$ K_b = \begin{cases}
K_0 + K_a, & \text{for } \Delta f < 0 \text{ (discharge)} \\
K_0 + K_a \cdot P_c, & \text{for } \Delta f > 0 \text{ (charge)}
\end{cases} $$
$$ T_b = \begin{cases}
T_0 + T_a, & \text{for } \Delta f < 0 \text{ (discharge)} \\
T_0 + T_a \cdot P_c, & \text{for } \Delta f > 0 \text{ (charge)}
\end{cases} $$
The adaptive components \(M_a\), \(K_a\), and \(T_a\) are proportional to the RoCoF, frequency deviation, and frequency acceleration, respectively:
$$ M_a = \alpha_M \cdot \left| \frac{d\Delta f}{dt} \right|, \quad K_a = \alpha_K \cdot |\Delta f|, \quad T_a = \alpha_T \cdot \left| \frac{d^2\Delta f}{dt^2} \right| $$
Where \(\alpha_M\), \(\alpha_K\), \(\alpha_T\) are positive tuning gains. This design means:
– A larger RoCoF automatically increases the inertia response.
– A larger frequency deviation automatically strengthens the droop response.
– A larger frequency acceleration automatically amplifies the trend response.
2.2 SOC-Based Constraint and Operational Zones:
The constraint factors \(P_c\) and \(P_d\) are crucial for protecting the battery. They are defined using a smooth logistic function to prevent abrupt power changes. The battery’s operational range is divided into zones to balance grid support with battery health.
| State | Zone | SOC Range | Constraint & Action |
|---|---|---|---|
| Discharge | Safe | SOC > 0.3 | \(P_d = 1\). Full adaptive power available. |
| Warning | 0.2 ≤ SOC ≤ 0.3 | \(P_d\) decays logistically from 1 to 0. Power is smoothly reduced. | |
| Danger | SOC < 0.2 | \(P_d = 0\). Discharge is blocked to prevent over-discharge. | |
| Charge | Safe | SOC < 0.7 | \(P_c = 1\). Full adaptive power available. |
| Warning | 0.7 ≤ SOC ≤ 0.8 | \(P_c\) decays logistically from 1 to 0. Power is smoothly reduced. | |
| Danger | SOC > 0.8 | \(P_c = 0\). Charge is blocked to prevent over-charge. | |
| Emergency (Grid) | \(|\Delta f| > 0.2\) Hz | Constraints are overridden. The cell energy storage system delivers/absorbs maximum power for urgent grid stability. | |
This zoning strategy ensures the cell energy storage system provides robust support when the battery is in a healthy state, gracefully scales back as limits are approached, and strictly enforces safety boundaries, all while prioritizing critical grid needs.
3. Hybrid Multi-Method Adaptive SOC Estimation
Accurate SOC is the linchpin for the aforementioned constraint strategy. Relying on simple Ampere-hour (Ah) integration leads to cumulative errors from current measurement inaccuracies and unknown capacity fade. We propose a hybrid, adaptive-weighting method that combines the short-term stability of Ah integration with the long-term accuracy of a model-based estimator (Unscented Kalman Filter – UKF).
3.1 Model and Methods:
The battery is modeled using a second-order RC equivalent circuit. The state-space model for the UKF is defined with states \(X = [SOC, U_1, U_2]^T\), where \(U_1, U_2\) are polarization voltages.
State Update:
$$ X_{k+1} = A \cdot X_k + B \cdot I_k + w_k $$
$$ A = \begin{bmatrix} 1 & 0 & 0 \\ 0 & e^{-\Delta t / (R_1 C_1)} & 0 \\ 0 & 0 & e^{-\Delta t / (R_2 C_2)} \end{bmatrix}, \quad B = \begin{bmatrix} -\eta \Delta t / Q_{max} \\ R_1(1 – e^{-\Delta t / (R_1 C_1)}) \\ R_2(1 – e^{-\Delta t / (R_2 C_2)}) \end{bmatrix} $$
Measurement (Voltage):
$$ V_{t,k} = OCV(SOC_k) – I_k R_0 – U_{1,k} – U_{2,k} + v_k $$
Where \(w_k\) and \(v_k\) are process and measurement noise.
The Ah-integration estimate is concurrently calculated, corrected for battery state-of-health (SOH):
$$ SOC_{Ah}(t) = SOC_0 – \frac{\eta}{Q_{max} \cdot SOH} \int_0^t I(\tau) d\tau $$
3.2 Adaptive Hybridization:
The final SOC estimate is a weighted average of the two methods. The key innovation is the adaptive weight adjustment based on the consistency of each method’s voltage prediction error.
Let:
– \(\Delta V_{Ah} = V_{t,measured} – OCV(SOC_{Ah})\)
– \(\Delta V_{UKF} = V_{t,measured} – OCV(SOC_{UKF})\)
The final SOC is calculated as:
$$ SOC_{final} = \beta \cdot SOC_{Ah} + \gamma \cdot SOC_{UKF} $$
with \(\beta + \gamma = 1\).
The weights \(\beta\) and \(\gamma\) are not fixed but adapt based on the sign agreement of the voltage errors:
– If \(\Delta V_{Ah}\) and \(\Delta V_{UKF}\) have opposite signs, it indicates the true SOC likely lies between the two estimates. The weights are balanced (\(\beta \approx \gamma\)).
– If \(\Delta V_{Ah}\) and \(\Delta V_{UKF}\) have the same sign, both methods are biased in the same direction. The weight shifts favorably towards the method historically deemed more reliable, typically increasing \(\gamma\) for the UKF over time as it converges.
A simple time-based adaptation rule can be: \(\gamma = min(1, t/t_{transition})\), where \(t_{transition}\) is a chosen time constant (e.g., 60 seconds), ensuring a smooth transition from reliance on Ah-integration at startup to reliance on the converged UKF estimator during steady operation.
| SOC Estimation Method | Maximum Error | Average Error | Key Characteristics |
|---|---|---|---|
| Ampere-Hour Integration | ~0.42% | ~0.21% | Simple, drifts over time, good initial accuracy. |
| Unscented Kalman Filter (UKF) | ~0.88% | ~0.13% | Accurate post-convergence, poor initial accuracy, computationally involved. |
| Proposed Hybrid Adaptive Method | ~0.09% | ~0.01% | Combines strengths: stable start, accurate steady-state. Optimal for cell energy storage system control. |
This hybrid approach ensures that the constraint logic for the cell energy storage system power output is always based on a highly reliable SOC value, fundamentally safeguarding the battery asset.
4. System Modeling and Performance Evaluation
The proposed strategy is evaluated using a Simulink model comprising a 1000 MW thermal generator (with governor and turbine dynamics) and a 30 MW / 10 MWh battery cell energy storage system. The thermal plant has a PFR deadband of ±0.033 Hz and an upper power limit of 60 MW for frequency response. The cell energy storage system has a smaller deadband of ±0.02 Hz and a response time constant \(T_E = 0.1s\). The system load damping factor \(D=1\) and inertia constant \(M=10\) s. The performance is compared against three benchmarks: (1) No storage, (2) Variable-K method (coefficients vary linearly with SOC), and (3) A conventional adaptive S-curve method (coefficients follow a logistic curve with SOC).
4.1 Case 1: Step Load Increase (60 MW)
A 60 MW step increase in load occurs at t=10s. The following table and analysis summarize the grid frequency performance.
| Control Strategy | Maximum Frequency Deviation (Hz) | Steady-State Frequency Deviation (Hz) |
|---|---|---|
| No Cell Energy Storage System | -0.096 | -0.063 |
| Variable-K Method | -0.082 | -0.058 |
| Conventional Adaptive Method | -0.067 | -0.053 |
| Proposed Comprehensive Strategy | -0.061 | -0.049 |
The results clearly demonstrate the superiority of the proposed method. By employing the adaptive tri-modal control, the cell energy storage system more effectively arrests the frequency drop (improved nadir) and contributes to a better steady-state settling point. The Trend Control specifically mitigates the initial sharp decline, while the adaptive coefficients ensure an optimal power response throughout the event.
4.2 Case 2: Continuous Random Load Fluctuations
A more realistic scenario with random load variations over 10 minutes is simulated, starting from an initial SOC of 0.5.
| Control Strategy | Avg. |Δf| (Hz) | Δf Peak-to-Peak (Hz) | Avg. SOC | SOC Peak-to-Peak |
|---|---|---|---|---|
| No Cell Energy Storage System | 0.048 | 0.197 | N/A | N/A |
| Variable-K Method | 0.044 | 0.170 | 0.490 | 0.021 |
| Conventional Adaptive Method | 0.041 | 0.140 | 0.482 | 0.037 |
| Proposed Comprehensive Strategy | 0.038 | 0.126 | 0.474 | 0.052 |
The proposed strategy achieves the lowest average frequency deviation and the smallest peak-to-peak swing, indicating superior overall frequency stabilization. The SOC deviation is larger because the strategy more actively utilizes the cell energy storage system‘s capacity to provide grid support, which is its intended function. Crucially, the hybrid SOC estimator ensures this occurs within safe bounds, preventing the SOC from entering the Danger zone.
Conclusion
This article has presented a comprehensive and adaptive control strategy to unlock the full potential of a battery-based cell energy storage system in providing primary frequency regulation. The strategy is built on a tri-modal control law incorporating trend, inertia, and droop responses, with all coefficients dynamically adapted based on real-time frequency dynamics (deviation, RoCoF, acceleration) and the internal state of the cell energy storage system. A sophisticated zoning logic based on a highly accurate, hybrid adaptive SOC estimation method protects the battery from harmful over-charge and over-discharge cycles.
Simulation studies under both step and continuous load disturbances confirm the strategy’s effectiveness. Compared to conventional methods, it significantly improves key frequency metrics—reducing the maximum frequency deviation by approximately 9-10% and the steady-state deviation by about 8% in a step disturbance scenario. It also provides the best overall frequency containment under fluctuating loads. The core achievement is the simultaneous optimization of grid performance and battery asset management, ensuring the cell energy storage system is a reliable, high-performing, and long-lasting resource for grid stability. Future work will focus on optimizing the weight adaptation law in the SOC estimator and extending the strategy to coordinate multiple, distributed cell energy storage system units.