Inertia in Converter-Dominated Microgrids: Control Strategies and Estimation Techniques

Abstract

This scoping review analyzes the role of inertia in converter-dominated microgrids, with an emphasis on hybrid AC/DC architectures. Following the PRISMA-ScR methodology, 54 studies published between 2015 and 2025 were identified, screened, and synthesized. The review addresses two key aspects, inertia estimation methods and control strategies for emulating inertia via power converters, emphasizing the role of the interlinking converter (ILC) as a bidirectional interface for inertia support between the AC and DC subsystems. This work addresses several limitations of prior reviews: their narrow scope, often overlooking advanced data-driven approaches such as machine learning; the lack of systematic classifications, hindering a comprehensive overview of existing methods; and the absence of practical guidance on selecting appropriate techniques for specific conditions. The findings show that conventional estimation methods are insufficient for low-inertia grids, necessitating adaptive and data-driven approaches. Virtual inertia emulation strategies—such as Virtual Synchronous Machines, Virtual Synchronous Generators, Synchronverters, and ILC-based controls—offer strong potential to enhance frequency stability but remain challenged by scalability, adaptability, and robustness. The review highlights critical research gaps and future directions to guide the development of resilient hybrid microgrid control strategies.

1. Introduction

Microgrids comprise DERs, controllable loads, and ESS that work in concert to ensure a reliable electricity supply and reduce energy costs [1]. According to IEEE Standard 1547–2018, a DER is an electricity generator, including RES, batteries, and power electronic converters, not connected to the bulk power system [2]. A microgrid can operate either connected to the main grid or independently in island mode, the latter being the primary focus of this study. To enable this flexibility, the microgrid must be able to transition between these two modes seamlessly. In grid-connected mode, the microgrid can provide ancillary services through commercial agreements governed by the regulations defining the utility–microgrid relationship [3,4,5]. However, providing reliable power is just one of many microgrid benefits; they also offer intelligent solutions for resolving conflicts among stakeholder interests, thereby enabling decisions aligned with market needs for optimal performance.

The deployment of DERs in power systems has grown significantly in recent years as part of global efforts to accelerate the energy transition, mitigate climate change, and address sustainability challenges. However, this shift presents a fundamental challenge: the reduction in system inertia due to the replacement of conventional power plants with DERs. Traditional synchronous generators, with their large rotating masses, store kinetic energy that provides system inertia, inherently counteracting frequency fluctuations. Furthermore, factors such as the intermittency of RES, the decentralized nature of distributed generation, and the growing penetration of power electronic loads like PHEVs introduce significant variability to the system’s overall inertia. This variability alters the distribution of inertia across the grid, resulting in areas with low inertia [6]. As system inertia decreases, the grid becomes vulnerable to large frequency deviations and a high ROCOF [7]. These conditions jeopardize power system stability and increase the risk of cascading failures or blackouts.

Inertia estimation methods are commonly classified based on their temporal application and the scope of analysis. From a temporal perspective, they can be grouped into post-mortem (offline), real-time (online), and forecast-based approaches. Regarding scope, estimation can be performed at different levels, including system-wide, regional, generation-side, demand-side, or at an individual resource level [8]. In power systems dominated by power electronic converters, wind and photovoltaic generation are among the most prominent renewable sources. However, the use of MPPT control prevents these technologies from providing active power support to mitigate frequency fluctuations [9,10]. DFIGs [11] are widely employed in wind power generation due to their ability to operate at variable speed and constant frequency. Although DFIGs inherently possess rotational inertia from wind turbine masses, this inertia is decoupled from the grid through the power converters [12]. As power systems become increasingly complex and dominated by converter-based resources, the reduction in system inertia poses new challenges for stability and control. In this context, developing reliable methods to estimate inertia has become a critical research priority. However, existing reviews on this subject are still limited in scope, often addressing only specific techniques without offering a comprehensive comparison or synthesis. Therefore, there is a clear need for an in-depth evaluation that organizes the available approaches, highlights their advantages and drawbacks, and provides guidance on their applicability in low-inertia microgrids.

Three primary approaches for virtual inertia emulation have been proposed: the VSM, the VSG, and the Synchronverter [13]. Among these, VSG control is commonly recommended for weak grids. In such networks, however, VSGs can exhibit performance limitations, particularly when implemented as current-source inverters. These limitations arise because the high variability of grid impedance directly impacts system stability, as the controller’s performance is highly dependent on network characteristics. Compensating for these effects requires estimating the grid inductance, a process that increases the design and operational complexity of these controllers [14]. These limitations highlight the need for more advanced controllers that provide scalability, adaptability, and self-learning capabilities. Traditional PI controllers are not well-suited for adapting to grid changes or learning from operating conditions. Therefore, developing intelligent controllers based on adaptive algorithms and modular structures is essential for enhancing system stability and enabling a more flexible and efficient integration of renewable energy sources into modern grids [15].

With the growing adoption of virtual inertia strategies in microgrid design, HMG configurations that integrate AC and DC subgrids exhibit distinct inertial behaviors. The DC link plays a crucial role as a dynamic indicator of the system’s energy state and can be utilized to stabilize the behavior of the AC network. In particular, variations in the DC voltage can be used to anticipate changes in generation or demand. Virtual inertia-based controllers enable smoother power transfer to the AC side. This control strategy considers not only the voltage deviation but also its rate of change, thereby introducing an adaptive damping mechanism to counteract sudden disturbances. Consequently, this approach prevents abrupt system responses and oscillations in the AC grid, enabling smooth and controlled power transitions. Thus, the DC link functions as a damper that filters disturbances, safeguarding the stability of the AC system [16]. Nevertheless, the direct interconnection of these heterogeneous subgrids does not inherently ensure an optimal dynamic response [17]. Consequently, developing control strategies to coordinate the inertial dynamics of both subgrids is essential for reducing both dynamic and steady-state deviations in HMGs and mitigating power oscillations at the ILC [18].

Regarding current inertia estimation methods, although they deliver reliable results in conventional systems with synchronous generators, their applicability is limited in converter-dominated networks. Among these, ROCOF-based approaches have been the most extensively studied. However, the literature still lacks comprehensive and comparative evaluations of their performance. In general, the existing studies tend to focus on a single estimation technique under specific conditions, which limits their applicability and generalization This gap is particularly critical in microgrids, which require a more holistic characterization of available estimation methods that accounts for their operational and structural diversity. In these variable environments, accurate inertia estimation is crucial for robust control [19].

This article investigates the implications of and remedies for low inertia in converter-dominated microgrids (MGs), including estimation techniques as well as control strategies. The article begins by reviewing the basic notion of inertia, highlighting its essential role in ensuring frequency stability during system disturbances. Particular emphasis is dedicated to hybrid microgrids that are composed of AC and DC sources, exhibiting distinct operational features. The second part discusses how inertia can be quantified in these systems—a task of growing relevance due to the increasing penetration of non-synchronous generation. The estimation of effective inertia in these systems is not trivial and requires accounting for the dynamic response of an inverter-based resource. The article then discusses various control techniques developed to provide virtual inertia by means of power converters. These methods seek to emulate the stabilizing action of conventional generators through sophisticated control strategies and the selective use of energy stored in the system (typically at the DC link). The paper concludes with a discussion on the principal technical, regulatory, and economic pitfalls that are still to be overcome. It also identifies several promising future research avenues. These include developing real-time estimation methods and designing more flexible, robust controllers. Furthermore, it highlights the need for coordinating multiple distributed energy resources to supplement system inertia as power networks continue to evolve.

2. Literature Review Methodology

This study was conducted as a scoping review, following the PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) guidelines to ensure a transparent process for identification, selection, and synthesis. The overall selection process is illustrated in Figure 1, which presents the PRISMA flow diagram used to structure and document the review methodology.

2.1. Databases and Search Strategy

The literature search was conducted using the following scientific databases: IEEE Xplore, Scopus and Web of Science. These databases were selected for their comprehensive coverage of engineering and energy-related literature. The search included articles published between 2015 and 2025. The primary search strings used included:

• “virtual inertia control AND hybrid microgrid”
• “inertia estimation” AND “power system”

2.2. Inclusion and Exclusion Criteria

Inclusion criteria: peer-reviewed journal articles and conference papers in English, addressing inertia estimation, inertia emulation, or control strategies in converter-dominated and/or hybrid AC/DC microgrids.

Exclusion criteria: studies published before 2015, studies published in languages other than English, studies that focus on microgrids and power system if they do not address specific frequency control and stability. The overall selection process and the number of studies at each stage are illustrated in Figure 2, which presents the PRISMA-ScR flow diagram. The PRISMA checklist is detailed in Appendix A.

2.3. Selection Process

Following an initial screening of titles and abstracts based on the defined criteria, 125 between papers, standards and books were selected for full-text review. From this set, a final selection of 54 studies was included in the scoping review.

2.4. Data Extraction and Synthesis

The selected studies were analyzed and categorized according to two main aspects:

▪

Inertia estimation methods (analytical, adaptive, statistical, AI-driven, frequency-domain).

▪

Inertia emulation and control strategies (Virtual Synchronous Machines, Virtual Synchronous Generators, Synchronverters, and interlinking converter-based controls in hybrid AC/DC microgrids).

The synthesis focused on identifying the applicability, advantages, limitations, and research gaps associated with each category, which guided the discussion and future research directions.

3. Results and Discussion

In low-inertia systems, primary frequency response becomes critical. Of particular interest is the frequency nadir, the minimum frequency reached after a disturbance before recovery begins, as it serves as a key indicator of system stability following events such as sudden generation loss. In low-inertia systems, the nadir can be reached within seconds, allowing for a rapid assessment of a disturbance’s potential impact on stable grid operation [20]. Predicting this nadir in advance would enable proactive corrective actions to suppress frequency excursions and prevent under-frequency load shedding.

Unlike conventional synchronous machines, Type IV wind turbines (full-converter-based) are completely decoupled from the grid frequency by their back-to-back converters, which enables variable-speed operation. While this maximizes energy extraction through MPPT, it also prevents them from naturally responding to frequency disturbances. In contrast, Type III turbines (DFIGs) use partial converters and have their stator directly connected to the grid, allowing limited interaction with frequency dynamics. Consequently, these converter-based systems do not inherently contribute to inertial support or participate in primary frequency regulation. In contrast to SGs, power converters lack an inherent dynamic response to frequency deviations, as they have no rotating mass or electromechanical coupling with the grid. Consequently, they do not naturally replicate the three primary mechanisms of frequency support: inertial response from kinetic energy stored in the rotor, primary control via speed and droop regulation, and secondary control through AGC [21]. Nevertheless, through advanced control algorithms—such as those based on VSM models or Synchronverter frameworks, virtual inertia can be synthesized by power electronic converters, enabling them to actively support system frequency stability [22]. This emulation relies on control strategies that replicate the inertial response of synchronous machines. Crucially, for these controllers to operate effectively, sufficient energy must be available at the DC-link—either stored in capacitors or supplied by connected energy storage systems. This ensures the converter can inject or absorb power during transient frequency events. Consequently, virtual inertia emulation has become a key technique to enhance the dynamic performance and resilience of inverter-dominated power systems [23,24,25,26].

The VSG strategy emulates the rotational inertia and damping characteristics of a synchronous generator through a power electronic inverter. This approach makes it applicable to photovoltaic systems, wind generation, and ESS for enhancing frequency support in microgrids. However, under different operating conditions, the inertia and damping support capabilities of the VSG can vary. The conventional VSG strategy typically employs fixed control coefficients, which can compromise dynamic regulation performance and reduce robustness [27]. Furthermore, the existing VSG strategies often lack clear guidelines for optimally tuning the inertia and damping coefficients. Additionally, because ROCOF is particularly susceptible to measurement noise, its use in defining control coefficients may produce disproportionate gain values, thereby increasing the risk of VSG instability [28].

3.1. Inertia Modeling and Frequency Response in Modern Microgrids

3.1.1. Inertia

During an imbalance between power generation and demand, different frequency response mechanisms are triggered across various time intervals, as illustrated in Figure 3. Following a disturbance, such as a sudden generation loss, the inertial response occurs, typically lasting less than 10 s. During this phase, the kinetic energy stored in SGs is released to counteract the energy shortfall, thereby limiting the frequency nadir and reducing the ROCOF. This is followed by the primary frequency response, which is generally activated within 10 to 30 s after the disturbance and relies on speed governor control to further mitigate frequency deviations. Finally, the secondary response, driven by AGC, acts within minutes to restore the system frequency to its nominal value [29].

The magnitude of this inertial response is quantified by the inertia constant, H, which represents the kinetic energy stored in a generator at synchronous speed relative to its rated apparent power. This constant is a characteristic parameter of each generator and is defined by (1).

$$H={\displaystyle \frac{1}{2·S_n}}J·w^2$$

(1)

Here, J denotes the generator’s moment of inertia (kg·m²), ω represents the nominal mechanical angular speed of the rotor (rad/s), and Sn is the rated apparent power of the generator (VA). System inertia is commonly described either as rotational kinetic energy or through the inertia constant, both of which serve as key indicators for evaluating frequency stability risks. The corresponding kinetic energy E_C is given by (2).

$$E_c=H·S_n$$

(2)

Accordingly, in a power system comprising n synchronous generators, the swing equation describing the dynamics of the ith generator can be written as (3).

$$2H_{SG,i}{\displaystyle \frac{d\Delta {\omega }_{g,i}}{dt}}=\Delta P_{m,i}-\Delta P_{e,i}-D_i\Delta {\omega }_{g,i}$$

(3)

where ΔP_m,i is the mechanical power variation of the ith generator, ΔP_e,i is the electrical power variation of the ith generator, Δω_g,i is the rotor speed deviation of the ith generator, and D_i is the damping factor of the ith generator.

In this context, damping represents the system’s capacity to suppress frequency oscillations following a disturbance, facilitating a faster and more stable return to the nominal frequency. Damping generally reduces both the magnitude and duration of frequency deviations caused by abrupt changes in generation or load. However, many virtual inertia schemes focus solely on emulating inertia while neglecting the damping component. This omission can be critical: without adequate damping, virtual inertia may reduce the initial frequency deviation but still allow for persistent oscillations, or even lead to system instability [30]. The inertia constant of power plants depends on the generation technology: hydroelectric units typically range from 1.75 to 4.75 s, nuclear plants around 4 s, and thermal plants from 2 to 10 s, influenced by size, turbine design, and operating conditions [22].

3.1.2. ROCOF

The ROCOF is a metric that represents the speed at which the electrical frequency changes in a power system. Mathematically, it corresponds to the first derivative of frequency with respect to time [31], and it can be expressed as (4).

$$ROCOF={\displaystyle \frac{∆P_d·f_O}{2·H·S}}={\displaystyle \frac{df}{dt}}$$

(4)

where f_o is the base frequency, ∆P_d is a function of the power imbalance, H the inertia constant, S is the base power

In microgrids, particularly those with high penetration of renewable energy sources such as solar or wind, the presence of synchronous generators is often limited or nonexistent, resulting in low overall system inertia. In this context, ROCOF becomes a critical indicator of the system’s dynamic state, as lower inertia leads to higher ROCOF values, reflecting greater sensitivity to disturbances.

Accurately estimating ROCOF following a disturbance is a key challenge. Traditional methods calculate ROCOF by numerically differentiating the measured frequency signal; however, this approach is susceptible to significant errors caused by measurement noise and signal artifacts post-disturbance. To overcome these limitations, some studies have proposed using the derivative of ROCOF, which can be estimated using Equation (5) [32].

$${\displaystyle \frac{d^{2}f}{d{t}^2}}={\displaystyle \frac{f_O}{2·H·S}}{\displaystyle \frac{d∆P_d}{dt}}$$

(5)

where f_o is the base frequency, ∆P_d is a function of the power imbalance, H is the inertia constant y S is the base power

The second derivative of frequency is used in inertia estimation because it is directly related to the system’s inertial behavior following a disturbance, as described by the swing equation for synchronous generators. In the initial moments following a disturbance, before primary control actions are activated, this variable primarily reflects the inertial response, allowing it to be isolated for real-time estimation. However, its direct use presents certain limitations, as values close to zero can lead to numerical [33].

3.1.3. Microgrids

The push for greater grid resilience amid rising renewable integration has made microgrids a cornerstone of the energy transition. These systems offer more than just local supply; they provide a framework for optimized energy management, streamlined integration of renewables, and reduced greenhouse gas emissions [34,35,36]. Whether connected to the main grid or operating in islanded mode, their flexibility is a key asset.

Theoretically, microgrids represent a departure from the centralized model, promising more efficient and sustainable energy management. They are enabled by the convergence of DERs, advances in power electronics, and automation technologies that allow for real-time control and optimization [37]. This has made them crucial for improving energy security, operating critical facilities, and enabling rural electrification. However, their full potential is constrained by a significant hurdle: developing advanced control strategies that can perform reliably in complex and unpredictable environments [38,39,40].

3.1.4. Hybrid Microgrids

Hybrid AC/DC networks represent an advanced microgrid architecture that combines AC and DC infrastructures to more efficiently integrate diverse distributed energy resources (DERs), such as solar, wind, and energy storage systems [41]. This configuration offers several key advantages. First, it improves energy efficiency by directly connecting native DC sources and loads, which significantly reduces power conversion losses and enhances overall system performance. Second, hybrid microgrids provide considerable operational flexibility, capable of functioning in both grid-connected and islanded modes to adapt to varying operational conditions. Finally, this architecture can lead to cost reductions by minimizing the need for intermediate converters, thereby lowering infrastructure, installation, and maintenance expenses. The coexistence of AC and DC buses allows for more efficient management of these resources, promoting a dynamic and optimized system balance.

However, HMG networks also pose certain challenges. One of the main disadvantages is the increased system complexity: the coexistence of two types of electrical buses results in a more intricate architecture, complicating the system design, operation, and maintenance. Additionally, stability issues may arise, as the coordination and synchronization between the AC and DC sections can be problematic, especially under variable load conditions or during faults [42]. Addressing these challenges requires more advanced control strategies to effectively manage the interaction between subsystems, which increases the technical complexity and implementation requirements [43]. Figure 4 illustrates the basic architecture of an HMG. It consists of both AC and DC subgrids, which are interconnected via an ILC. This converter enables coordinated power exchange between the subgrids while playing a key role in the control of frequency and voltage stability. Various DERs, such as photovoltaic arrays, wind turbines, diesel generators, and ESS, are connected to the respective buses, along with AC and DC loads.

3.1.5. Role of Inertia Estimation in Modern Power System Dynamics

With the rapid growth of renewable generation, inertia estimation has shifted from an academic exercise to a critical operational need. An accurate estimate enables operators to determine whether the system can withstand the variability of intermittent sources and, in turn, adapt their strategies to prevent failures. In practice, this information is fundamental for optimizing resources: from designing more efficient controls to adjusting reserve margins to improve the system’s response to disturbances.

Moreover, inertia estimation plays an important role in system-level analysis and coordination of synthetic inertia contributions, even though some converter-based technologies, such as Synchronverters, do not directly depend on estimated inertia for their operation. Instead, these systems rely on predefined control parameters to emulate a virtual inertial response.

Traditional inertia assessment methods are often complex, with certain steps relying on highly idealized assumptions. Processing large volumes of operational data in modern power systems can be time-consuming, which limits the applicability of these methods in real-world scenarios. Most existing inertia prediction methods are based on data-fitting techniques or the use of neural networks. However, data-fitting approaches require extensive reference datasets and involve exhaustive scanning of the data space during computation. While BP neural networks offer strong nonlinear approximation and learning capabilities, they tend to exhibit slow convergence speeds and are prone to settling in local minima, which reduces prediction efficiency. The current trend involves combining predictive approaches, which requires large volumes of operational or historical inertia data for reference [44].

The integration of machine learning and artificial intelligence, leveraging their advanced data processing capabilities, holds significant potential for improving inertia prediction. However, while machine learning is well established in many fields, its application to inertia forecasting is still in its early stages. This is partly due to the unique characteristics of power systems with high levels of renewable energy integration, where inertia behaves differently from that of conventional systems. The spatial and temporal variability of renewable generation further complicates the system’s dynamic behavior, making inertia prediction more challenging. These factors complicate the estimation overall system inertia and the development of effective training datasets, posing significant challenges for applying machine learning in modern energy systems.

3.2. Inertia Estimation Methods

In recent years, the energy transition has significantly reshaped the dynamics of power systems. This transition involves the progressive disconnection of conventional synchronous generators—which traditionally provide physical inertia—and the increasing integration of non-synchronous renewable energy sources, leading to a substantial reduction in total system inertia. This decline has resulted in higher ROCOF values, thereby compromising the system’s stability and response capability during disturbances. In this new context, traditional inertia estimation methods, typically based on the swing equation and assumptions of linear and predictable system behavior, have lost accuracy. Consequently, there is a growing need to revise, adapt, and classify current estimation methods to improve their accuracy in modern power systems. This requires evaluating a wide range of techniques—from analytical and frequency-domain approaches to machine learning—for their effectiveness in environments characterized by low rotational inertia and high renewable penetration.

Various schemes have been proposed for classifying inertia estimation methods. One common approach is to categorize them by their temporal application: post-mortem (offline) methods, real-time (online) methods, and forecasting approaches [19]. Post-mortem methods are typically used in analytical scenarios to retrospectively assess the level of inertia during significant disturbance events. Real-time methods [45] aim to provide instantaneous inertia estimates using readily available system measurements. Forecasting approaches, on the other hand, estimate the expected inertia at future time intervals. A summary of these methods is presented in

Table 1. Classification of Inertia Estimation Methods.

Estimation Method	References	Advantages	Disadvantages
ARMAX-based estimation (statistical)	[46,47]	Robustness in systems with high penetration of renewables.	Computationally intensive, especially in large or highly variable systems.
SWM (adaptative)	[46,48]	Provides continuous inertia estimations.	High computational demand.
DMD (model-based)	[46,49]	Can estimate inertia using ambient data.	Accuracy may decrease in environments with a poor signal-to-noise ratio.
Models based on ROCOF and the swing equation (analytical)	[8,50,51]	Provides rapid inertia estimations.	Numerical instabilities may arise.
Post-mortem offline (measured data)	[8]	Low computational demand.	The accuracy of estimations can be affected by the magnitude of the disturbance.
Estimations via PMU measurements (analytical)	[49,50]	They provide synchronized and high-frequency data, which improves accuracy.	The continuous processing of large volumes of data requires advanced infrastructure and algorithms.
Continuous methods (statistical)	[8]	Provides a continuous stream of estimations.	It is difficult to accurately estimate power imbalances during normal operations.
Predictive methods (AI-based)	[46]	Significantly aids in operational planning.	Accuracy is heavily dependent on the quality of the input data.
Zonal or area-based estimation (analytical)	[8]	Allows for estimating inertia in large zones, key in networks where it is not feasible to track every machine.	Each area combines generators and inverters and unifying them into a single model introduces uncertainty and reduces precision.

.

Another way to classify these methods is based on how the variable of interest is obtained: they can be broadly categorized into model-based approaches and measurement-based approaches, as illustrated in Figure 5.

3.2.1. Analytical Methods

Analytical methods for inertia estimation rely on mathematical formulations rooted in power system physics. These methods are particularly effective in systems dominated by synchronous generators, whose dynamic response is accurately described by established equations. The classical approach relies on the swing equation, which relates the power imbalance to frequency variation, allowing system inertia to be inferred. Generally, at least two parameters must be estimated to ensure proper control system performance, making synthetic inertia estimation challenging due to the difficulty of obtaining the required system states. Some progress has been made in this area through the use of extended EKF [52].

Within this analytical category, one notable method is polynomial fitting-based estimation. Using LSM, the inertia constant of a synchronous generator can be determined under large-scale transient conditions. Simulations have demonstrated high accuracy over time windows of several seconds and under various operating conditions, including steady-state operation [53]. This approach provides valuable insights into the availability of inertia for future inertia markets. Moreover, the LSM technique can be enhanced using methods such as Tikhonov regularization or Lasso regression [54]. A recursive approach has also been proposed, utilizing active power and frequency measurements from PMUs. After preprocessing to improve measurement accuracy, the rotor speed is estimated independently of the system model. A recursive least squares formulation is then applied in the z-domain based on the equation error, with results subsequently converted to the s-domain. This method is noted for its precision and low computational load [55].

An alternative method derives inertia from the initial ROCOF observed after a generation–load mismatch, leveraging the inverse relationship between ROCOF and inertia. To enhance reliability, the approach employs a flexible-window least squares algorithm together with a median filter, enabling accurate real-time ROCOF estimation under high levels of renewable integration [56]. The traditional polynomial fitting method for inertia estimation in power systems generally relies on a fixed-order polynomial. This method is applied to the frequency response recorded after a known disturbance, fitting a curve—typically a fifth-order polynomial—to the frequency trajectory to capture the generators’ inertial behavior. Nevertheless, the accuracy of this approach is affected by factors such as system size, network topology, and the magnitude and location of the disturbance. To address these limitations, some studies have proposed variable-order polynomial fitting to more effectively represent the frequency response. Additionally, polynomial fitting has been applied to total power generation to estimate the disturbance magnitude, which is not always directly measurable. These refinements aim to more accurately capture the system’s inertial response within a short time window, thus improving estimation accuracy over conventional methods [57].

3.2.2. Adaptive Methods

Adaptive inertia estimation methods dynamically adjust their parameters in response to varying power system conditions. Strategies such as the SWM process real-time data sequences to compute moving averages and obtain continuous, accurate estimates, making them particularly useful in environments with a high penetration of intermittent renewables. Other approaches, such as the R, V, and RV methods [58] incorporate load models that are sensitive to frequency and voltage variations, thereby improving accuracy in systems with dynamic loads. While these methods are valued for their adaptability and precision, they require high-quality data and can be susceptible to measurement noise.

A lesser-known technique involves the use of multivariate O-splines, which are differentiators based on a specific type of filter applied to the DTTFT. These differentiators are called O-splines because their segments are spaced in intervals of one fundamental frequency cycle. O-splines are used to precisely compute ROCOF and active power deviations in power systems. They are integrated into the real-time processing of synchronized signals, enabling fast and accurate estimation of inertia constants in generating units. These filters provide the best approximation of Taylor–Fourier coefficients through a matrix that expands the Fourier subharmonic space by incorporating higher-order Taylor terms. WAMS [59] are used in conjunction with O-splines to estimate inertia without relying on system models, using measurements of active power deviation and frequency after load changes or generator disconnections [60].

3.2.3. Statistics Methods

Statistical methods for inertia estimation rely on data analysis without requiring detailed physical models and are primarily divided into two approaches. On one hand, stochastic models capture the variability and uncertainty of systems with high renewable energy penetration using processes such as Markov or Gaussian models [61], enabling accurate estimates even under normal operating conditions or minor disturbances, without the need for major events. However, their application demands significant statistical expertise, substantial computational resources, and high-quality data to ensure reliable results.

On the other hand, the ARMAX model [47,62], employs regressions of historical frequency and active power data, along with noise terms, to predict frequency deviations. In medium-sized systems with high renewable penetration, this method proves particularly useful. Its robustness and suitability for real-time operation make it attractive, yet its reliance on wide data windows and precise parameter tuning poses practical challenges. A promising alternative is the construction of a multi-input, multi-output ARMAX model that integrates both inertial and primary responses. This model is built using ambient frequency and active power deviations and can estimate system inertia by analyzing data from normal operation as well as transient events (ringdown). By deriving transfer functions that describe the relationship between active power and frequency, the method enables the calculation of the inertia constant, which is crucial for accurately capturing power system dynamics [45]. Similarly to ARIMA [63] in that it uses past values and past errors, ARMAX adds an additional component, exogenous variables, which are external factors that influence the behavior of the time series.

Another method discussed in the literature is the ARX model, used for online inertia estimation in power systems. This approach models the system as an ARX structure, where the inputs are active power increments and the output is the frequency deviation, since inertia cannot be directly observed as a measurable variable during system operation. The ARX model is characterized by having two polynomials, one for the outputs and one for the inputs, along with a delay operator. In this process, the model parameters are estimated in real-time and adjusted using both traditional estimation methods and the DREM technique. The latter is particularly useful because it enables accurate and fast identification of ARX model parameters even under system “noise” conditions [64]. The DREM method works by extending and mixing dynamic regressors to improve parameter estimation in nonlinear systems. The key idea is to reformulate the estimation problem into a structure that allows the use of stable linear operators to generate new variables with enhanced regressors. This approach decomposes the original estimation problem into smaller and more manageable sub-problems, enabling the accurate estimation of multiple parameters [65].

Some real-time forecasting tools have also been implemented and have shown promising results, predicting system inertia and primary frequency response up to three hours in advance [19]. To improve forecasting accuracy in renewable energy systems, the ARIMA model has been applied, as it can handle the uncertainty associated with wind and solar power. This model predicts the system’s kinetic energy in the short term, avoiding the issues of referencing inertia to a common base, as encountered in other approaches [66]. Although the ARIMA model provides meaningful results, further improvements have been achieved in real systems using a Bayesian model that incorporates MCMC sampling and a limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) algorithm for optimization. This model forecasts 30 min sequences of the power system’s kinetic energy. The advantage lies in the fact that the Bayesian approach accounts for uncertainty through a predictive distribution, which the ARIMA model does not [63].

3.2.4. Model-Based Methods

Model-based inertia estimation methods require an explicit mathematical model of the system, in which inertia is estimated as an unknown parameter. These methods can be broadly categorized into three main groups. The first group includes approaches based on the swing equation, which estimate inertia using frequency and power measurements. These methods are valued for their simplicity and widespread use but have limitations stemming from their reliance on synchronous generator models, assumptions of ideal dynamics, and sensitivity to measurement noise [57,67]. To overcome these limitations, researchers have proposed modified versions of the swing equation that incorporate damping terms and avoid divisions by near-zero derivatives, thereby improving accuracy in systems with more complex dynamics.

The second group comprises the KF [68] and its variants, such as the EKF and the UKF. These algorithms estimate the system state, including inertia, in real time by refining predictions through successive updates with measured data. While they are robust to noise and effective for dynamic monitoring, their implementation requires an accurate dynamic model and can involve significant computational overhead.

Finally, the third group includes model-aided identification methods, ref. [69] being a notable example. This technique involves injecting low-amplitude signals and fitting a first-order model to the observed response. It does not require large disturbance events, making it suitable for heterogeneous systems with high renewable penetration. However, it demands careful design of the excitation signal and may be sensitive to noise. In summary, methods based on the swing equation are simple but require large disturbances; Kalman filter-based methods provide greater operational precision at the cost of increased complexity; and micro-perturbation techniques offer a versatile alternative for modern systems without relying on major events.

The HSE method based on the ISHUKF has the potential to support system inertia estimation. This is because the model used includes frequency estimation as part of the state variables, and system inertia is closely related to the system’s ability to resist frequency changes caused by load or generation variations. Moreover, the ISHUKF is capable of handling dynamic noise and frequency deviations, making it suitable for assessing system dynamics, particularly in low-inertia conditions as found in systems with high integration of renewable sources [70].

When power systems experience major disturbances, the power imbalance during the inertial response phase is dominated by the deviation in electrical power. This occurs because the generator’s governor takes longer to react compared to the inertial response; consequently, mechanical power remains in a steady state or shows minimal changes during the inertial process. The frequency deviation of each generator in a power system is often evaluated with respect to the frequency of the COI, which plays an important role in power system stability analysis and wide-area control. The COI represents the inertia-weighted average motion of all synchronous generators in the electrical system. The frequency derived from the COI is smoother than local measurements, reflects the overall system behavior, and filters out electromechanical oscillations. These properties make it useful for control systems, especially for inertia emulation, improving frequency response without compromising system stability [71]. In practice, the COI frequency is typically calculated during dynamic simulation stage of the power system. This process can be time-consuming for large-scale systems and, therefore, is often not performed online. However, with the deployment of PMUs across wide areas, the online calculation of COI frequency becomes feasible [72].

The COI refers to the common, uniform motion shared by all synchronous machines in a power system. It represents the system-wide frequency, which may not correlate with local electromechanical dynamics of individual generators. Some authors have proposed using the COI for inertia emulation, highlighting that its frequency exhibits low observability of electromechanical modes, thereby improving the response to frequency transients without compromising system stability [47,73]. The eigenstructure-based method analyzes a power system’s dynamics by first representing it with a state-space model. By computing the eigenvalues and eigenvectors from synchronized measurements (e.g., frequency and power at specific nodes), the system’s dominant modes can be identified. The method then analyzes the dynamic relationships within these modes between speed and power deviations to estimate the effective system inertia. This data-driven approach allows for inertia estimation without requiring traditional generator-specific information [49]. Dynamic equivalents (DEs) can reduce computational effort and highlight the main characteristics of a power system. While some model parameters, including inertia, are not identifiable using only the steady-state data before and during a disturbance they become identifiable when using post-disturbance data; among these parameters is inertia [74].

3.2.5. Artificial Intelligence Based Methods

Inertia estimation is increasingly shifting toward artificial intelligence, which can infer system dynamics from minor fluctuations in real time, sidestepping the need for major disturbances. This trend is dominated by the use of ANNs in their various forms, from foundational architectures to more advanced variants like CNNs and RNNs that promise greater accuracy for dynamic network management [75,76,77]. These models leverage historical PMU data—capturing variables like active power and frequency—to estimate both inertia and the magnitude of system events. Other promising techniques, like the SINDy method, can construct dynamic models from scratch by analyzing key measurements, making them valuable in [78].

The appeal of these methods lies in their flexibility and ability to learn from vast datasets. However, these methods also present significant challenges. Their effectiveness is entirely dependent on access to high-quality, synchronized data and considerable computational power. Furthermore, they introduce the risk of overfitting and are notoriously difficult to interpret, often operating as a “black box” that obscures the physical reasoning behind their estimations—a critical drawback in a field where explainability is paramount for trust and safety.

The estimation phase is commonly based on an ANN trained using algorithms such as Levenberg–Marquardt, which allows for real-time calculation of the system’s inertia with high accuracy. This network constitutes the core of the proposed approach, due to its ability to adapt to dynamic conditions and its robustness against noise, making it an ideal tool for grids with high-RES penetration. The complete process also includes the collection of key variables such as voltage, rotor angle, rotor speed, and maximum mechanical power; the identification of optimal fault points using the Fault Location Index (FLI); and the application of the Maximum Equal Area Criterion (MEAC) to determine the critical clearing angle. This integration of advanced techniques enables a reliable and efficient estimation of [79]. Estimating inertia can be achieved by employing convolutional neural networks (CNNs), trained with phase voltage and frequency signals recorded by PMUs [50,80]. An alternative method estimates equivalent inertia using a Gaussian Markov model to capture the hidden relationship between average frequency variations and system inertia. Additionally, PSO-SVM models have been employed, using power variation and frequency rate as inputs to improve prediction accuracy. Results show that PSO-SVM significantly outperforms GA-BP neural networks and conventional SVM, reducing prediction errors by up to 23.64% and 68.27%, respectively, and proving effective in tracking inertia variations during load changes [81]. In addition to data-driven approaches such as Gaussian Markov models and PSO-SVM, the Predict-and-Optimize (PAO) method applied in Finland combines probabilistic forecasting with robust optimization to predict daily inertia and optimize it, thereby reducing operational costs and risks [82].

3.2.6. Frequency-Domain-Based Methods

Frequency-domain methods for inertia estimation operate in the spectral domain, using tools such as the DFT [83] and the Wavelet Transform [84,85], to analyze the system’s response to disturbances. The Wavelet Transform, in particular, offers advantages by providing both temporal and frequency information simultaneously, making it especially useful in systems with fast or non-stationary dynamics. The method proposed in [86] combines the injection of test signals by a Grid-Forming Converter Interfaced Generator (GF CIG), the analysis of the system response in the frequency domain, and data-fitting techniques to efficiently estimate the global inertia. Since this method relies solely on measured data, it becomes independent of the specific electrical characteristics of the system and is thus capable of estimating global inertia.

3.2.7. Data-Driven Methods

These methods rely on the direct use of information obtained from the electrical power system, either in real time or from historical records. This category includes techniques that utilize advanced devices such as PMUs, which enable the capture of electrical variables with high temporal resolution and precise synchronization. It also encompasses the use of historical time series data obtained from SCADA systems or event logs, which can be processed to estimate frequency variations over time. These approaches do not necessarily depend on complex mathematical models of the system; instead, they focus on observing the electrical behavior through actual measured data. In some cases, the raw measurement data is processed using statistical techniques, machine learning, or adaptive filtering to improve the accuracy and robustness of the inertia estimation. Their main advantage lies in the ability to provide inertia estimates that more closely reflect the real behavior of the system, particularly in highly dynamic environments with significant penetration of renewable energy sources.

3.3. Inertia Control

Low-inertia power systems are vulnerable to rapid frequency oscillations, which may result in frequency dips, disconnections, or even blackouts. To mitigate these risks, virtual inertia methods have been developed using controlled inverters [87], energy storage systems [88], and other technologies [89] capable of emulating the inertial behavior of traditional synchronous generators.

In microgrids, inertia sources can be diverse. Conventional inertia is provided by synchronous generators such as diesel engines, whose rotating components contribute mechanical inertia (with a typical constant of H = 3.5 s) [90]. In contrast, virtual inertia can be provided by electronic systems such as voltage source converters (VSCs), grid-connected electric vehicles, data centers, and lithium-ion batteries. With proper control strategies, these devices can be controlled to emulate the behavior of synchronous generators. In hybrid AC/DC systems, the physical nature of inertia differs by subgrid type: in AC systems, it is related to the moment of inertia of rotating masses, while in DC systems, it is analogous to the total capacitance of connected sources. Coordinated use of these technologies enhances dynamic stability [91], increases system resilience, and ensures reliable operation, even under high renewable penetration and variable demand.

The implementation of virtual inertia control has become a key strategy to enhance frequency stability in systems with increasing integration of renewable energy sources. This technique is realized through various technological solutions, with the VSG [92] being one of the most representative. VSGs emulate the dynamic behavior of conventional synchronous generators through inverters, allowing simultaneous control of both frequency and voltage, and often use LCL filters to mitigate harmonics. Additionally, the ESS, such as batteries and supercapacitors, provide fast frequency response, and hybrid configurations are capable of effectively managing both fast and sustained events. To optimize the performance of these systems, advanced control methods are integrated, including derivative control, H∞ control, PID controllers, fuzzy logic, and neural networks, all of which improve system adaptability and robustness [93]. The implementation process involves measuring electrical magnitudes, synchronizing with PLLs, calculating control signals, and tuning parameters such as inertia (J) and damping coefficient (D). Key design aspects such as proper storage sizing, coordination with schemes like droop control [94], and small-signal stability analysis are essential to ensure efficient and stable operation of the power system under dynamic conditions.

3.3.1. Inertia in AC Microgrids

To address the frequency instability caused by CIG, VI) has become the main line of defense. Although the idea of implementing VIE through control strategies in converters is well-established, its precise quantification within the total system inertia remains a surprisingly underexplored topic in the literature [12]. It is crucial to understand that virtual inertia is a control technique, not a physical energy reserve. Its primary objective is to reduce the ROCOF, by providing an initial damping effect, rather than to arrest the frequency nadir or preserve natural inertia. The BESS contribute to frequency stability in low-inertia microgrids by emulating inertia, while supercapacitors provide fast frequency support in islanded systems due to their rapid energy delivery [95]. Some of the main techniques proposed for this purpose are illustrated in Figure 6. It is acknowledged that other types of ESS, such as EVs, flywheels, and others, were identified during the literature search. However, these technologies fall outside the defined scope of this review and were therefore not considered in the analysis. The focus was limited to a representative subset of ESS technologies directly related to inertia support and control in hybrid microgrids.

Although the frequency nadir can be stabilized by fast frequency reserves or primary frequency response, an inertia response con also be provided by fast-acting energy sources connected to the grid, as long as they are adequately controlled via power electronic converters. Inertia emulation is typically achieved by configuring renewable energy converters to mimic the dynamics of synchronous machines, by enabling fast-response sources to inject power during disturbances, or by maintaining sufficient synchronous machine capacity in the grid to balance the increasing penetration of renewable generation. In combination with power electronics–based emulation strategies, deloading can be applied to wind turbine generators as well as photovoltaic systems, offering an effective means of enhancing system support [96]. In this approach, the generator is deliberately operated below its maximum available power (i.e., in a deloaded or curtailed state) to maintain a power reserve. This reserve can be rapidly deployed to compensate for system imbalances immediately following a disturbance, using fast-response measurement and control techniques. For a PV system in deloaded mode, its inverter’s operating point is set below the maximum power point (MPP). The unit remains connected to the grid, continuously monitoring frequency and ROCOF. Upon detecting a disturbance, the control system rapidly adjusts the operating point toward the MPP to inject the reserved power into the grid.

In the case of wind turbines, deloading is achieved by operating them at a rotational speed slightly above the optimal MPPT level. In this state, the turbine is already injecting power into the grid but maintains a reserve of kinetic energy. During a disturbance, the turbine injects additional power by allowing its speed to decrease toward the MPPT level. During the power injection phase, the turbine’s speed gradually decreases until it reaches the MPPT level, thereby eliminating recovery periods and providing continuous energy support, removing the need for primary frequency reserves.

A control strategy that can be employed is the bang-bang control, based on binary logic that switches between two extreme states, similar to an on-off switch, without intermediate values. In the context of frequency control in microgrids, this traditional strategy adjusts the virtual inertia (J) to its maximum (J_max) or minimum (J_min) values depending on the frequency deviation (Δf) and its rate of change (df/dt). J_min and J_max are determined through system stability and dynamic response analysis to ensure both fast response and stability under disturbances. When both parameters share the same sign, indicating a deviation away from the nominal frequency, the inertia is increased to slow down the oscillation. Conversely, when the signs are opposite, signaling a return to nominal, the inertia is reduced to speed up recovery. However, this logic can lead to constant fluctuations in the inertia in response to minor disturbances, triggering unnecessary control actions that may compromise system stability. To address this issue, an improved version of the Bang-Bang strategy have been proposed, incorporating a steady-state interval, a tolerance zone within which the inertia remains fixed at a stable value (J_s), thus reducing oversensitivity and avoiding unnecessary oscillations [97].

Among the available integration options, DC microgrids are particularly attractive due to their high efficiency and ease of interfacing with various DERs, including energy storage. When connected to the AC grid, they use a DC-AC converter as the interface. However, they still face significant challenges for full integration into traditional power systems [98]. Despite these challenges, DC microgrids offer important benefits, such as seamless DER integration, improved frequency regulation, and enhanced decentralized droop control strategies that lead to better power management. Some authors classify non-synchronous inertia sources into categories such as virtual inertia sources, primary responses from regulators, and voltage-/frequency-dependent loads [99].

3.3.2. Inertia in AC-DC Microgrids

Control strategies in hybrid AC–DC microgrids are generally categorized into local control and coordinated control. Local control operates at the device level without requiring external communication. Coordinated control, on the other hand, includes three main approaches: (i) centralized control, where optimal decisions are made by a central unit but rely heavily on communication infrastructure; (ii) decentralized control, which offers high reliability without communication links but may lead to suboptimal decisions; and (iii) distributed control, where decisions are made collaboratively among nodes, balancing efficiency and robustness. Additionally, a hierarchical control scheme is often proposed, composed of three levels: (i) primary control, which is fast and local; (ii) secondary control, which restores nominal values; and (iii) tertiary control, which handles energy optimization. Conventional and modified droop control methods are commonly employed for power management in these systems [100].

In hybrid microgrids (HMGs), which integrate AC and DC subsystems to optimize energy management, advanced control strategies have been proposed to ensure system stability and efficiency. One notable approach is the integration of VI [101] as part of the control scheme for ESS. This control strategy combines the SoC-based management with the VI concept to regulate charge/discharge operations, enhance dynamic response, and maintain system stability under load and generation fluctuations. The virtual inertia model used is based on a mathematical representation that emulates the dynamic behavior of physical inertia found in traditional synchronous generators. This model is typically described by the transfer function (6).

$$G_{JD}(s)={\displaystyle \frac{1}{J·s+D}}$$

(6)

where J represents the inertia coefficient, determining the system’s capacity to withstand sudden frequency changes, and D is the damping factor, which dissipates oscillations and contributes to system stabilization. Known as the virtual inertia oscillator model, it is widely adopted in energy control systems to improve transient response and system stability. Proper tuning of J and D is critical: increasing J enhances the system’s disturbance absorption but may reduce stability margins by shifting dominant poles closer to the origin; conversely, increasing D improves stability by pushing poles further from the origin but can limit the system’s responsiveness to fast changes. Thus, a trade-off is necessary to achieve adequate dynamic performance.

In the system proposed in [102], the VI model is integrated into the ESS’s Power Management System (PMS) controller, enabling the ESS to react dynamically to generation and load variations. This not only helps maintain frequency within acceptable bounds but also extends ESS lifespan by limiting charge/discharge rates, resulting in a more efficient and resilient system under renewable intermittency and load variability.

An advanced approach involves distributed coordination using Virtual Synchronous Generators (VSGs). This strategy includes (i) coordination of Net Power Control (NPC) to balance energy sharing between AC and DC subgrids, (ii) use of VSG-based inverters and Interlinking Converters (ILCs) to inject virtual inertia and improve dynamic response, (iii) distributed secondary control to restore voltage and frequency without centralized communication, and (iv) an Inertia Deviation Index to strategically allocate virtual inertia and mitigate oscillations. Advantages include accurate power sharing among all generators, simultaneous restoration of voltage and frequency without steady-state error, inertia support across subgrids, robust operation even in weak networks, fault tolerance in communication, and high scalability. However, challenges remain, such as the complexity of tuning critical parameters [103].

Additionally, the ROCOX control strategy, based on adaptive droop control using the ROCOF and ROCOV, has been proposed for ILCs. It aims to minimize frequency and voltage deviations, reduce power oscillations, and maintain ROCOF and ROCOV within safe limits. Benefits include enhanced dynamic response, automatic tuning of the droop coefficient K_d to balance AC and DC subgrids, and significant reduction in dynamic deviation indexes, validated through HIL environments. Nevertheless, limitations include the complexity of adaptive controller design, stability issues if K_d is improperly adjusted, and the hardware constraints of the ILC [104].

Among the most commonly used approaches is adaptive control, applied to the ILC in AC/DC hybrid microgrids. This technique dynamically adjusts virtual inertia and coupling coefficients based on ROCOF, enhancing transient system performance without affecting steady-state operation. It effectively reduces frequency overshoot, increases dynamic inertia, and supports bidirectional power flow between AC and DC subgrids, thus promoting global stability. However, it requires complex implementation, dependency on specific parameters, and may lead to DC voltage deviations during transients [105,106].

The use ILCs becomes particularly relevant when managing large volumes of energy, which often requires deploying multiple ILCs in parallel to distribute the load and avoid overburdening a single device. However, this configuration introduces technical challenges, such as the emergence of circulating currents caused by line impedance mismatches. These unwanted currents can compromise both the efficiency and reliability of the system. To mitigate these effects and ensure proper coordination among the converters, advanced control strategies are employed. These include (i) internal current control in the dq0 reference frame, (ii) normalized droop control for frequency and voltage regulation, and (iii) decoupling schemes based on the DC subgrid voltage. Additionally, an external control loop is implemented to manage power interaction between subgrids and ensure balanced current sharing among the ILCs [107].

3.4. Discussion and Future Work

Ensuring stability in modern power systems hinges on the critical task of inertia estimation, a challenge magnified by the proliferation of converter-based resources within microgrids. This structural shift inherently lowers system inertia, making the grid more sensitive to disturbances and prone to higher ROCOF events that jeopardize frequency stability [108]. The problem is compounded by practical measurement limits [109], including noisy data and incomplete observability across all nodes, which hinders accurate calculation. A further layer of complexity comes from the mixed response profiles: the instantaneous reaction of synchronous machines versus the delayed, control-dependent response of virtual inertia sources.

A key challenge lies in decoupling the contributions from these synchronous and non-synchronous sources, as traditional swing-equation models are ill-equipped for the complex dynamics of low-inertia networks. This makes it essential not only to estimate the total system inertia but also to map its spatial distribution at a nodal level. Such insight would allow operators to pinpoint vulnerable areas and make strategic operational decisions. Recent analytical work on estimating nodal inertia from steady-state parameters shows a promising path forward to address this limitation and could inform the optimal allocation of inertial resources in the future [110].

In hybrid microgrids, the analogy between the energy in a DC-side capacitor $\left({\displaystyle \frac{1}{2}}C{v}^2\right)$ and the kinetic energy in a rotating mass $\left({\displaystyle \frac{1}{2}}J{\omega }^2\right)$ provides the conceptual foundation for virtual inertia. While traditional generators provide this response naturally, converter-based systems must emulate it by leveraging the stored energy in their DC links to resist rapid frequency changes. This equivalence is reflected in the system’s governing swing equations, where the DC bus voltage and its time derivative become proxies for frequency and its rate of change.

However, implementing these control strategies introduces its own set of challenges, including the risk of instability from excessive virtual inertia or oscillations caused by control delays. Looking ahead, the central task must be large-scale experimental validation under real-world conditions to move beyond simulation. This will force the field to confront the most pressing questions: how do we automate parameter optimization with AI without creating unstable “black box” systems? And how can we guarantee the cybersecurity of the distributed control links that will manage the grid’s dynamic inertia?

Future research on inertia in converter-dominated and hybrid AC/DC microgrids should begin with a more objective evaluation of existing estimation methods. This requires detailed assessments of key performance metrics such as accuracy and computational complexity, supported by systematic comparisons through simulations under different operating conditions. Further studies should focus on optimizing the sizing and costs of energy storage systems to determine the best trade-offs between technical performance and economic feasibility in microgrids. Another important direction is the advancement of grid-forming inverters, addressing their large-scale integration, the development of innovative control schemes, and the evaluation of their technical and economic impacts.

In addition, coordinated control strategies between storage systems, ILCs, and renewable resources should be explored to provide robust and scalable virtual inertia support. Finally, the design of real-time and data-driven approaches, particularly those based on artificial intelligence, is essential for developing adaptive solutions that can effectively respond to the dynamic conditions of low-inertia microgrids. Together, these directions form a research agenda aimed at strengthening the stability, resilience, and efficiency of future power systems with high penetration of renewable energy sources.

4. Conclusions

This scoping review provided a structured synthesis of inertia in converter-dominated microgrids, with particular emphasis on hybrid AC/DC architectures. Using the PRISMA-ScR methodology, 54 studies published between 2015 and 2025 were identified, screened, and analyzed, focusing on two main aspects: inertia estimation methods and virtual inertia control strategies.

The findings confirm that conventional estimation methods, originally developed for synchronous generator systems, are insufficient for low-inertia microgrids. Adaptive and data-driven techniques, including machine learning approaches, have emerged as promising alternatives, yet they still require systematic validation under diverse operating scenarios. Similarly, virtual inertia emulation strategies—such as Virtual Synchronous Machines, Virtual Synchronous Generators, Synchronverters, and interlinking converter (ILC)-based controls—demonstrate strong potential to improve frequency stability in hybrid microgrids. However, their effectiveness continues to be limited by issues of scalability, robustness, and coordination among distributed resources.

An additional insight concerns the dynamic interaction between inertia and damping. While inertia shapes the immediate response to a disturbance, damping determines how oscillations evolve over time. In traditional power systems, damping played a secondary role in the short-term response, but in converter-dominated grids with high renewable penetration, both parameters must be jointly estimated and controlled to ensure stability.

This review also highlighted several research gaps that remain unaddressed in the current literature: the lack of systematic classifications of estimation and control methods, the insufficient benchmarking of existing techniques using objective performance metrics, and the absence of practical guidance on the conditions under which each approach is most effective. Further work is needed to establish standardized evaluation frameworks, optimize the integration of storage and ILC-based solutions, and design intelligent, adaptive, and real-time strategies to strengthen the stability, resilience, and efficiency of hybrid AC/DC microgrids.