In modern power grids, the increasing integration of renewable energy sources, such as wind and solar, has introduced significant challenges to frequency stability. The inherent intermittency and volatility of these sources reduce the system’s inertia and damping, making frequency regulation more complex and demanding. Traditional fossil-fuel-based generators often lack the rapid response capabilities required to counteract these fluctuations, leading to potential frequency deviations that can compromise grid reliability. In this context, battery energy storage systems have emerged as a critical solution due to their fast response times, bidirectional power flow, and flexibility in providing ancillary services. This article delves into the application of battery energy storage systems for primary frequency modulation, focusing on a data-driven, model-free adaptive control strategy that enhances grid adaptability and performance without relying on precise mathematical models. The battery energy storage system is pivotal in stabilizing frequency, especially in high-renewable-penetration scenarios, and this work explores its optimized control through innovative approaches.
The core of frequency regulation lies in maintaining a balance between generation and load. When a disturbance occurs, such as a sudden load change or renewable output variation, the system frequency deviates from its nominal value (e.g., 50 Hz or 60 Hz). Primary frequency modulation aims to quickly counteract these deviations using resources like governors and, increasingly, battery energy storage systems. The battery energy storage system can inject or absorb power within milliseconds, providing immediate frequency support. However, effective control of the battery energy storage system must consider multiple factors, including the state of charge (SOC) to ensure longevity, the grid’s frequency requirements, and the uncertainty of system dynamics. Traditional control methods, such as fixed or variable droop control, often depend on accurate models of the battery energy storage system and the grid, which can be difficult to obtain in practice. To address this, a model-free adaptive control (MFAC) approach is proposed, leveraging only input-output data from the battery energy storage system to dynamically adjust its output. This strategy not only improves frequency stability but also maintains the SOC within an optimal range, enhancing the overall efficiency and lifespan of the battery energy storage system.
To understand the impact of a battery energy storage system on frequency modulation, we first analyze the frequency characteristics of a regional power grid with high renewable penetration. The system model aggregates components like traditional generators, renewable sources, loads, and the battery energy storage system. For a grid without renewables, the frequency response to a load disturbance can be represented by a simple transfer function. Let $$\Delta P$$ denote the load disturbance, and $$\Delta f$$ denote the frequency deviation. The relationship is given by:
$$G(s) = \frac{\Delta f}{\Delta P} = \frac{1}{Ms + D}$$
where $$M = 2H$$ is the inertia constant (with $$H$$ as the inertia time constant), $$D$$ is the damping constant, and $$s$$ is the Laplace operator. This model assumes a linearized system around an operating point. When renewable energy sources are integrated, with a penetration rate $$p$$ (e.g., 20% of generation from renewables), the effective inertia decreases because renewables like wind and solar typically contribute little to system inertia. The modified inertia constant becomes:
$$M’ = (1 – p)M$$
Thus, the frequency transfer function changes to:
$$G'(s) = \frac{\Delta f}{\Delta P} = \frac{1}{M’s + D}$$
This reduction in inertia exacerbates frequency deviations, making the grid more susceptible to disturbances. To mitigate this, a battery energy storage system is incorporated into the model. The battery energy storage system provides additional power output $$\Delta P_b(s)$$ in response to frequency deviations, modeled as a first-order lag system with time constant $$T_b$$. Similarly, traditional generators contribute $$\Delta P_g(s)$$ based on their governor and turbine dynamics. The overall system model, including the battery energy storage system, can be represented as shown in the block diagram (not depicted due to image restrictions, but described mathematically). The transfer function from load disturbance to frequency deviation becomes:
$$\Delta F(s) = \frac{\Delta P_L(s)}{M’s + D + G_s(s)K_G + G_b(s)K_b}$$
where $$G_s(s)$$ is the transfer function of the traditional generator (including governor, turbine, and reheater dynamics), $$G_b(s) = \frac{1}{1 + sT_b}$$ is the transfer function of the battery energy storage system, $$K_G$$ is the generator’s regulation gain, and $$K_b$$ is the battery energy storage system’s regulation gain. The frequency response characteristics can be analyzed using Bode plots to assess amplitude-frequency behavior. For instance, with parameters like $$K_b = 24$$ (per unit), $$K_G = 20$$ (per unit), $$T_b = 0.1$$ s, and generator time constants $$T_G = 0.08$$ s, $$T_{CH} = 0.3$$ s, $$T_{RH} = 10$$ s, the amplitude-frequency function is:
$$H(s)_b = \frac{1}{(M’s + D) + G_s(s)K_G + G_b(s)K_b}$$
Comparatively, without the battery energy storage system, the function is:
$$H(s) = \frac{1}{M’s + D + G_s(s)K_G}$$
Analysis shows that in low-frequency ranges (e.g., below 0.2 Hz), the amplitude with the battery energy storage system is significantly lower than without it, indicating improved frequency stability. The battery energy storage system effectively dampens oscillations caused by renewable fluctuations and load changes, reducing both the maximum frequency deviation and the rate of frequency change. This underscores the importance of the battery energy storage system in enhancing grid resilience. The following table summarizes key parameters used in the frequency characteristic analysis:
| Parameter | Symbol | Value (per unit) | Description |
|---|---|---|---|
| Inertia Constant | M | 5 | Base inertia without renewables |
| Damping Constant | D | 1 | System damping coefficient |
| Battery Time Constant | T_b | 0.1 s | Response lag of battery energy storage system |
| Generator Regulation Gain | K_G | 20 | Droop control gain of traditional generator |
| Battery Regulation Gain | K_b | 24 | Droop control gain of battery energy storage system |
| Renewable Penetration | p | 0.1 to 0.2 | Fraction of generation from renewables |
The model-free adaptive control strategy for the battery energy storage system is designed to overcome the limitations of model-dependent approaches. MFAC is a data-driven method that uses only the input-output data of the controlled system—here, the battery energy storage system—to design the controller. It does not require an explicit mathematical model of the battery energy storage system or the grid, making it highly adaptable to uncertainties and changes in system dynamics. The core idea is to dynamically linearize the nonlinear system around each operating point using a pseudo-gradient vector, which is estimated from real-time data. For a single-input single-output (SISO) system, let the controlled system be represented by $$y(k) = f(u(k))$$, where $$y(k)$$ is the output (e.g., frequency deviation or SOC), $$u(k)$$ is the input (e.g., power adjustment from the battery energy storage system), and $$f$$ is an unknown nonlinear function. At each time step $$k$$, a compact form dynamic linearization model is constructed:
$$\Delta y(k+1) = \phi(k) \Delta u(k)$$
where $$\Delta y(k+1) = y(k+1) – y(k)$$, $$\Delta u(k) = u(k) – u(k-1)$$, and $$\phi(k)$$ is the pseudo-gradient, a time-varying parameter that captures the system’s dynamics. The pseudo-gradient is estimated recursively using the following algorithm:
$$\hat{\phi}(k) = \hat{\phi}(k-1) + \frac{\eta_k \Delta u(k-1)}{\mu + |\Delta u(k-1)|^2} \left[ \Delta y(k) – \hat{\phi}(k-1) \Delta u(k-1) \right]$$
where $$\hat{\phi}(k)$$ is the estimate of $$\phi(k)$$, $$\eta_k$$ is a step-size parameter, and $$\mu > 0$$ is a weighting factor to avoid division by zero. The control law is then derived to minimize the tracking error between the desired output $$y_r(k+1)$$ and the actual output $$y(k+1)$$. The control input is computed as:
$$u(k) = u(k-1) + \frac{\rho \hat{\phi}(k)}{\lambda + |\hat{\phi}(k)|^2} \left[ y_r(k+1) – y(k) \right]$$
where $$\rho$$ is a learning factor (typically in (0,2]), and $$\lambda > 0$$ is a penalty factor to constrain the control input magnitude. This iterative process continuously updates the pseudo-gradient and control input based on new data, enabling the battery energy storage system to adapt to changing grid conditions. For primary frequency modulation, the battery energy storage system’s control system is designed as a multiple-input single-output (MISO) system, with inputs being the grid frequency deviation $$\Delta f$$ and the battery SOC $$Q_{SOC}$$, and the output being the power adjustment $$\Delta P_b$$. The desired output $$y_r$$ is set to zero for frequency deviation (aiming to restore frequency to nominal) and to a reference SOC (e.g., 0.5) for SOC management. The controller dynamically balances these objectives, ensuring that the battery energy storage system provides rapid frequency support while maintaining SOC within a safe range (e.g., 0.2 to 0.8) to prolong battery life. The following table outlines the MFAC algorithm parameters and their roles:
| Parameter | Symbol | Typical Range | Purpose |
|---|---|---|---|
| Step-size | η_k | 0.01 to 1 | Adjusts estimation update rate |
| Weighting Factor | μ | 0.001 to 0.1 | Prevents denominator singularity |
| Learning Factor | ρ | 0.1 to 1.5 | Controls input update aggressiveness |
| Penalty Factor | λ | 0.1 to 1 | Limits control input magnitude |
To evaluate the performance of the battery energy storage system under the proposed MFAC strategy, we define two sets of metrics based on the type of load disturbance: step disturbances and continuous disturbances. For step disturbances, key indicators include the maximum frequency deviation $$|\Delta f_m|$$, the minimum frequency deviation $$|\Delta f_s|$$, the time to reach maximum deviation $$t_m$$, the time to reach steady-state $$t_s$$, the frequency drop rate $$\alpha$$, and the recovery rate $$\beta$$. These are calculated as:
$$\alpha = \frac{|\Delta f_m|}{t_m}, \quad \beta = \frac{|\Delta f_m| – |\Delta f_s|}{t_s – t_m}$$
A smaller $$|\Delta f_m|$$ and $$|\Delta f_s|$$ indicate better frequency stability, a smaller $$\alpha$$ signifies stronger disturbance rejection, and a larger $$\beta$$ implies faster recovery. For continuous disturbances, metrics focus on the dispersion of frequency and SOC from their nominal values over time. The frequency index $$Q_f$$ and SOC index $$P_{SOC}$$ are defined as:
$$Q_f = \sqrt{\frac{1}{N} \sum_{i=1}^{N} (f_i – f_0)^2}, \quad P_{SOC} = \sqrt{\frac{1}{N} \sum_{i=1}^{N} (Q_{SOC_i} – Q_{SOC_0})^2}$$
where $$f_0 = 50$$ Hz (or 60 Hz) is the nominal frequency, $$Q_{SOC_0} = 0.5$$ is the target SOC, and $$N$$ is the number of samples. Lower values of $$Q_f$$ and $$P_{SOC}$$ denote better performance in maintaining frequency and SOC, respectively. These metrics allow for a comprehensive assessment of the battery energy storage system’s effectiveness in primary frequency modulation across different scenarios.
Simulation studies are conducted using Matlab/Simulink to validate the proposed MFAC strategy for the battery energy storage system. A regional grid model is built with parameters as described earlier, including a traditional generator (100 MW rated capacity) and a battery energy storage system (1 MW/1 MWh capacity). Four control strategies are compared: (1) no battery energy storage system participation, (2) fixed droop control (termed “fixed-K”), (3) variable droop control based on SOC (termed “variable-K”), and (4) the proposed MFAC strategy. The simulations consider two disturbance types: step load changes and continuous load variations. For step disturbances, a load increase of 0.08 per unit is applied at time t=0 s. The frequency deviation responses are recorded, along with SOC trajectories. Under initial SOC of 0.5, the MFAC strategy yields a maximum frequency deviation of 0.1861 Hz, comparable to fixed-K (0.1861 Hz), but with better SOC maintenance. The recovery rate $$\beta$$ is 0.0034 for MFAC, higher than fixed-K’s 0.0021, indicating quicker stabilization. When initial SOC is lower (0.3), fixed-K causes early battery depletion, leading to secondary frequency dips, while MFAC maintains performance without such issues. The following table summarizes the step disturbance results for initial SOC=0.5:
| Control Strategy | |Δf_m| (Hz) | |Δf_s| (Hz) | α (Hz/s) | β (Hz/s) | SOC Final |
|---|---|---|---|---|---|
| No BESS | 0.4715 | 0.2262 | 0.4710 | 0.0067 | N/A |
| Fixed-K | 0.1861 | 0.1231 | 0.4834 | 0.0021 | ~0.4 |
| Variable-K | 0.2510 | 0.1589 | 0.3643 | 0.0024 | ~0.45 |
| MFAC | 0.1861 | 0.1218 | 0.4834 | 0.0034 | ~0.48 |
For continuous disturbances, two scenarios are simulated: short-term (5-minute) and long-term (1-hour) load fluctuations. The load profile consists of random variations around a mean, representing realistic renewable and demand changes. In the 5-minute case, the MFAC strategy achieves a frequency index $$Q_f = 0.0959$$, slightly higher than fixed-K’s 0.0901 but lower than variable-K’s 0.1182, demonstrating competitive frequency stabilization. The SOC index $$P_{SOC}$$ is 0.0804 for MFAC, indicating good SOC preservation. In the 1-hour scenario, MFAC excels with $$Q_f = 0.1432$$ and $$P_{SOC} = 0.2530$$, outperforming fixed-K ($$Q_f = 0.1496$$, $$P_{SOC} = 0.4485$$) and variable-K ($$Q_f = 0.1489$$, $$P_{SOC} = 0.2561$$). The battery energy storage system under MFAC avoids deep discharge and maintains SOC closer to the target, ensuring sustained availability for frequency modulation. The results highlight the adaptability of the MFAC strategy across different time scales and disturbance patterns. Additional simulations with varying renewable penetration levels (e.g., p=0.1 to 0.3) show that MFAC consistently reduces frequency deviations by 20-30% compared to no battery energy storage system, while keeping SOC within safe bounds. The table below compares the continuous disturbance metrics for the 1-hour case:
| Control Strategy | Q_f (Hz) | P_SOC | Remarks |
|---|---|---|---|
| No BESS | 0.1605 | N/A | High frequency volatility |
| Fixed-K | 0.1496 | 0.4485 | Poor SOC maintenance |
| Variable-K | 0.1489 | 0.2561 | Moderate performance |
| MFAC | 0.1432 | 0.2530 | Best overall balance |
The superiority of the MFAC strategy stems from its data-driven nature, which allows the battery energy storage system to continuously learn and adapt to grid conditions without requiring a priori knowledge of system parameters. Unlike fixed-K, which applies constant gains regardless of SOC, or variable-K, which switches modes based on thresholds, MFAC dynamically adjusts control actions using real-time feedback on both frequency and SOC. This minimizes model dependency and enhances robustness against uncertainties, such as changes in renewable output or load patterns. Moreover, the MFAC algorithm’s computational simplicity ensures it can be implemented in real-time control systems for battery energy storage systems, making it practical for field applications. The battery energy storage system thus becomes a more intelligent and resilient asset for grid frequency support.
In conclusion, this article presents a comprehensive study on the use of battery energy storage systems for primary frequency modulation, with a focus on a model-free adaptive control strategy. The analysis of frequency characteristics confirms that integrating a battery energy storage system significantly improves grid stability, especially in high-renewable-penetration scenarios. The proposed MFAC strategy leverages input-output data to dynamically control the battery energy storage system, effectively balancing frequency regulation and SOC management. Simulation results demonstrate that MFAC outperforms traditional fixed and variable droop controls in terms of frequency deviation reduction and SOC preservation, particularly under continuous disturbances. The battery energy storage system under MFAC exhibits enhanced adaptability, robustness, and longevity, contributing to a more reliable and sustainable power grid. Future work could explore the integration of multiple battery energy storage systems in larger networks, the combination of MFAC with other advanced control techniques, and real-world validation through pilot projects. As the energy transition accelerates, data-driven approaches like MFAC will be crucial for optimizing the performance of battery energy storage systems and achieving carbon neutrality goals. The battery energy storage system, empowered by adaptive control, stands as a cornerstone of modern grid resilience and efficiency.