UFLS and Smart Load for Frequency Regulation in Electrical Power System: A Review

The frequency of an electrical power system can be interpreted as an indicator of the equilibrium between generation and demand. Under fault conditions where this balance is disrupted, the Under-Frequency Load Shedding scheme is typically the ultimate resource employed to prevent the frequency from dropping to undesirable thresholds that could lead to a blackout. Smart loads have the capability to dynamically adjust their consumption in response to frequency variations, which has the potential to enhance frequency regulation and mitigate the need for abrupt load disconnection through the Under-Frequency Load Shedding scheme. The objective of this paper is to present a literature review of Under-Frequency Load Shedding schemes, an innovative approach to their classifications, and the introduction of a new category within these schemes that incorporates smart loads. Additionally, an overview of smart loads is included to be potentially employed in the design of new schemes and contribute to a shift in the operation philosophy, considering that the primary mission of electric power systems is to provide energy to loads rather than disconnecting them.


I. INTRODUCTION
In order to maintain frequency within defined ranges that do not affect the operation of the power system or threaten the integrity of its elements, automatic load shedding schemes known as Under Frequency Load Shedding (UFLS) have been used.These schemes ensure that in the event of sudden faults that cause imbalances between demand and generation (affecting frequency and voltage stability [1]), the system can maintain its stability and return the frequency to suitable operating values.Frequency can be understood as an index of the balance between demand and generation for electric power systems operating in alternating current.Therefore, when the balance between demand and generation is interrupted and the latter is lower than the former, the system frequency begins to decrease [2], [3].Conversely, The associate editor coordinating the review of this manuscript and approving it for publication was Fabio Mottola .
if generation exceeds demand, the system frequency tends to increase.
From the moment the power imbalance between generation and demand occurs until the critical point for the system is reached, there is a period of time in which actions or switching must be carried out to restore power balance and frequency stability [4].It is during this aforementioned time that UFLS can operate in order to prevent system malfunctions, instabilities, undesired islands, and blackouts resulting from frequency disturbances.To achieve this goal, an UFLS performs the gradual disconnection of loads until balance is reached [4], reducing the harmful effects of blackouts on the network [5].
For years, in traditional electrical power systems where synchronous and rotating machines, in general, were the main source of supply, the levels of inertia made the dynamic response slow.Therefore, when there was a variation in load or generation levels, the energy needed to cover the initial FIGURE 1. Behavior of frequency with loss of inertia.The dotted curve represents the frequency behavior with the loss of a generator in the IEEE 39 Busbar test system, considering the total system inertia.The solid curve shows the frequency behavior with the loss of 50% of the total system inertia.
imbalance was naturally released from the stored kinetic energy [6] in the rotating masses, allowing adequate time to perform control actions and restore generation-demand balance [7], [8].However, with the decrease in inertia levels nowadays, the nadir reached during a frequency event and the rate of change of frequency (RoCoF) tend to increase, as shown in Figure 1 that presents frequency as a function of time.When there is an imbalance between generation and load, this is compensated for by the system inertia H s , represented in equation (1), and the inertia of each generator H ( g, i), as in equation (2) [7].
where E kin,i is the kinetic energy of the generator, J is the moment of inertia, S s is the total power of the system, f n is the nominal frequency, and S g,i is the power of each generator.
With the integration of elements connected to the electrical grid through power electronic interfaces, the total inertia of the system decreases, as can be seen in equation (1), resulting in faster and less damped frequency responses, making the system less stable, as shown in Figure 1, where the inertia decreases by 50% and the frequency nadir increases.Traditional UFLS schemes have been designed with the consideration of electrical power systems with appropriate levels of inertia, based mainly on rotating machines.Therefore, in the face of variations inherent in new-generation systems, there is a need to create schemes to preserve frequency stability and overall system stability, including actions that improve UFLS performance or reduce the need for its operation.
According to the Institute of Electrical and Electronics Engineers (IEEE) standards, load shedding due to subfrequency should be performed quickly enough to stop the frequency decrease and balance the generation with demand [9].To maintain the frequency at adequate levels, a dead band or zone is established (in which the UFLS does not operate), limited by threshold values that activate the UFLS when a power imbalance representing a risk to the system is detected.In turn, the acceptable frequency nadir varies from one system to another depending on the type of generators, auxiliary devices, and turbine governors [4].In this way, the UFLS is triggered when the system exceeds a frequency value and disconnects specified load quantities in steps or defined portions [10].
With the modernization of electrical power systems, Phasor Measurement Units (PMUs) have been implemented and can be used to monitor frequency and RoCoF in real-time with high precision and sampling speeds.The UFLS can be activated based on control commands from a centralized entity or control center, depending on the values obtained from the PMUs [11].Due to their wide acceptance and advances in measurement systems, it is possible to achieve an easier migration from traditional UFLS to more robust, smart, flexible, and adaptable schemes [4].
Currently, the philosophy of operation of UFLS is focused on disconnecting portions of loads when frequency instability or active power deficits are detected.This is the last protective action taken by the system when an unsafe frequency condition is detected.In modern networks where inertia levels are affected by the connection of non-inertial generation, the frequency response to disturbances will be less damped with a higher RoCoF, and it is possible that the response of traditional UFLS schemes will become insufficient or result in excessive shedding of load to address the imbalance.
In the scenario of inertial loss, it is necessary to implement complementary schemes to UFLS to improve the system's response speed when imbalances occur between generation and demand.In this case, smart loads that have power electronics interfaces can quickly modify their consumption and could be understood as a type of UFLS system that changes its active power consumption in response to variations in frequency levels, providing inertial response and primary frequency regulation services.So, smart loads could improve the system's dynamics, reducing the impacts of traditional UFLS actions or even avoiding their operation.The categories of frequency regulation are presented in Figure 2, which details the response times of smart loads and traditional UFLS.
Inertia loss could limit renewable integration or increase the need for system strength by thermal or fossil fuel power plants due to the need to provide frequency regulation and inertial response services.As a result, [12] proposes a new restriction on the oscillation equation to include RoCoF and frequency nadir by employing UFLS services.In [13], UFLS, inertia, and restrictions on thermal power plants are employed to avoid major load shedding.On the other hand, [14] proposes a dynamically controlled optimization to choose between inertia, fast frequency regulation, and rapid response in order to avoid UFLS action.In these cases, UFLS is considered one of the reserves available to the grid operator to ensure frequency in an optimal and cost-effective manner [12].
In recent years, there have been some reviews of UFLS techniques and schemes.In [15], the proposal is to classify automatic UFLS schemes into conventional and advanced categories.Conventional schemes are classified as static and semi-adaptive, whereas advanced schemes are classified into computational intelligence, adaptive relay parameters, adaptive amount, stability, adaptive distribution, hybrid, and predictors.On the other hand, [4] suggests a similar classification in which UFLS schemes are divided into conventional UFLS, adaptive UFLS, computational techniques UFLS, and Wide Area Measurement System (WAMS) UFLS.These reviews did not consider smart loads as inputs for frequency regulation, and UFLS schemes based on smart loads were not taken into account.
Smart loads set a new milestone that can radically modify the concept of load shedding due to low frequency or the philosophy of operation of UFLS schemes.By modulating energy consumption rather than performing direct load shedding, it is possible to avoid reaching dangerous frequency values for the power grid.This was not possible in the past due to the characteristics of the loads that made up the electrical network.With new technologies and smart loads, new UFLS schemes and new philosophies are now possible.
UFLS scheme simulations are usually performed in three domains: time-domain simulations [16], [17], frequency response simulations [18], [19], and Monte Carlo simulations [20], [21].In time-domain simulations, the power system is modeled as a set of differential equations, and the UFLS scheme is implemented as a controller that adjusts the load shedding schedule based on the system frequency.In frequency response simulations, the power system is modeled as a set of transfer functions, and the UFLS scheme is implemented as a controller that adjusts the load shedding schedule based on the system frequency response.In Monte Carlo simulations, a large number of random scenarios are generated, and the UFLS scheme is evaluated in each scenario.This allows the performance of the UFLS scheme to be evaluated under a variety of conditions.
The first motivation for the present work is to address the remarkable need to comprehensively examine the emerging synergy between traditional UFLS methods and smart load management strategies.The fusion of these approaches holds the potential not only to prevent frequency disturbances but also to optimize the operation of the electrical grid by harnessing the inherent flexibility within connected loads.A rigorous analysis of this convergence would provide a deeper understanding of how energy systems can benefit from a holistic approach that leverages both active load response and advanced control capabilities.
This survey presents a literature review based on UFLS schemes developed by multiple authors.A new category is included for UFLS schemes based on smart loads.We also provide a review of smart loads, or loads connected through power electronics, that can provide frequency regulation services and reduce or prevent the action of UFLS and abrupt load shedding.
In section II, a concise description of the methodology employed for the literature review is presented.
The remainder of this paper is organized as follows.Section II present a concise description of the methodology employed for the literature review.Section III reviews various types of UFLS schemes.Section IV reviews various types of strategies for UFLS schemes and frequency regulation with smart loads and their classification.Section V discusses the main advantages and disadvantages of UFLS techniques.Section VI concludes this paper and provides suggestions for future work.

II. METHOD OF REVIEW
The purpose of this review is to aggregate available data, perform an in-depth analysis, and deliver a comprehensive assessment of modeling techniques concerning UFLS schemes and smart loads technology.This effort entails systematically exploring relevant literature.The pertinent references were collected and accurately referenced, encompassing materials from global platforms.The process of identifying pertinent publications involved utilizing suitable keywords associated with UFLS schemes and smart load applications for frequency regulation within electrical power systems.Numerous journals, conference papers, and book chapters were tackled during the review process.However, the selection of studies was guided by factors such as relevance, most cited, originality, abstract, key words, and contributions.
The existing reviews and classifications of UFLS schemes were examined.Subsequently, publications related to UFLS schemes were sought and then selected based on their content to identify a novel scheme.Once identified, a fresh classification of UFLS schemes was put forward, introducing the novelty of incorporating smart loads into this study.Ultimately, a new classification of UFLS schemes, including smart loads, was proposed.The general flow chart of the review work is presented in Figure 3.

III. UNDER FREQUENCY LOAD SHEDDING
To prevent frequency drops in power systems, one of the most important preventive measures is load reduction or shedding using an automatic system that reacts to frequency changes [11], [20], [22].The use of UFLS systems reduces the risk of uncontrolled separation of areas, loss of generation, and blackouts in cases of significant disturbances (such as the loss and disconnection of a large amount of generation) [23].
Automatic load shedding schemes for low frequency conditions can be classified as conventional or multi-stage, semi-automatic, and adaptive according to [11] and [24], or as conventional, adaptive, semi-automatic, calculationbased, and WAMS-based UFLS techniques, as proposed in [4].In addition, some authors have proposed UFLS techniques considering parameters other than frequency or RoCoF, for example, [25] presents a predictive method based on monitoring synchronous capacitor active power 110970 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.injections.In [26], power flows through transmission lines are considered constraints for UFLS operation.
The multi-stage UFLS scheme is considered the most widely used and applied in large power systems around the world due to its simplicity.Although with the implementation of renewable resources and power-electronics-connected devices, their operation is expected to become more resilient and may be displaced by adaptive UFLS schemes [24].In the multi-stage UFLS scheme, protection relays are activated by disconnecting defined portions of the load when the network frequency drops and remains below certain thresholds for a set delay time [27].The details of some multi-stage UFLS schemes implemented by utility companies are presented in [28], and in [20], uncertainties in generation deficiencies, inertia, and damping factor are ignored, and the Monte-Carlo method is used for developing a probabilistic UFLS scheme.Semi-adaptive UFLS schemes operate based on the frequency magnitude and RoCoF, adjusting their response based on the imbalance between demand and generation or the magnitude of the contingency [29].On the other hand, adaptive schemes focus on making modifications to load levels based on RoCoF fluctuations, so WAMS and PMUs are typically employed for their development as proposed in [30] and [31].
The constant variations of RoCoF make it difficult to implement adaptive UFLS schemes, so it is not common to find purely adaptive UFLS systems in practice [24].On the other hand, multistage UFLS schemes are usually the most common due to their simplicity.Although with the implementation of renewable resources and powerelectronics-connected devices, their operation is expected to become more resilient and may be displaced by adaptive UFLS schemes [24].In Figure 5, a simplified classification of the main UFLS techniques is presented.This includes (i) traditional UFLS schemes, (ii) adaptive schemes, (iii) schemes based on intelligent computational techniques, (iv) schemes based on WAMS, and finally (v) schemes based on smart loads.

A. TRADITIONAL UFLS
This method is the most commonly applied in power systems and serves as the basis for other UFLS schemes [4].It is based on the local measurement of frequency, which is compared with previously defined values [15].When the frequency reaches these values, the scheme disconnects predefined portions of load according to the frequency threshold reached [10].For these schemes, a minimum frequency threshold is defined, which is determined by the nominal system frequency and typically by the minimum turbine operating frequency [32].
Figure 4 shows the basic operation of a traditional UFLS scheme for a 50 Hz power system based on the ENTSO-E philosophy [33].The UFLS scheme starts operating when the frequency drops to 49 Hz, in this case a portion of the load equivalent to 5% is disconnected.If the system fails to balance generation and demand and the frequency drops to 48.8 Hz, another 5% of the load is disconnected.Once these two thresholds are exceeded, the system will continue to disconnect portions of the load, but this time in 10% fractions until reaching a total disconnection of 50% of the load at 47.8 Hz.Since each power system has its own characteristics compared to other systems, it has its own frequency thresholds based mainly on installed capacity or generation, types of generation, system inertia, and reserves, among other characteristics [4].
Several authors have proposed different conventional UFLS schemes.In [34], a method is presented based on a fixed and random priority combination of loads that gives to the system some flexibility and allows it to optimize resources.The presented UFLS is based on an 8-stage deghosting scheme.The initial de-ghosting starts at 49.5 Hz, and the last stage is activated at 48.5 Hz.In [35], several traditional UFLS schemes are presented based solely on frequency thresholds and some semi-adaptive schemes that complement frequency with RoCoF.Among the analyzed schemes, the Italian scheme stands out, which starts its operation at 49.1 Hz and reaches a 60% load shedding at the last threshold, which corresponds to 47.5 Hz.On the other hand, some authors are concerned about other variables that affect or are affected by load shedding, so in [36], the voltage stability limit is considered within the proposed UFLS scheme, and in [37], the load importance factor (LIF), the reciprocal phase angle sensitivity (RPAS), and the voltage electrical distance (VED) are considered.
As mentioned earlier, traditional or static multi-stage UFLS schemes are widely used around the world for their simplicity and efficiency.However, because they have finite load shedding steps, they can result in excessive or inadequate load shedding.To address this, [38] proposes a multi-stage UFLS method to determine the portion and location of disconnected loads.In [16], a probabilistic method was developed to design traditional multi-stage UFLS schemes considering system uncertainties such as generation outages, hourly load, wind speed, and frequency regulation provided by wind turbines.On the other hand, [39] proposes the use of an UFLS method based on a linear regression model developed by offline simulations using a set of representative cases.

B. ADAPTATIVE UFLS
The increased integration of new generation and demand technologies using power electronics interfaces do not provide, in general, mechanical inertia to the grid.Such feature makes the dynamics and response to frequency imbalances faster and more prone to instabilities.This raises the need for new UFLS schemes since fast frequency dynamics make traditional UFLS unable to detect initial imbalances between generation and demand to prevent the collapse of the power system [40].In [41], the development of an UFLS that considers the variation of inertia levels in the power grid is considered.The proposed scheme estimates the RoCoF, the center of inertia (CoI), and the size of the generation loss (LoG) using only local frequency measurements.Additionally, a novel technique for detecting the inflection point is presented to eliminate the effect of local frequency oscillations.On the other hand, [40] proposes a method for reconfiguring conventional UFLS to adapt to the changing dynamics of the power system.
Active power imbalances also lead to mismatches in reactive power, which can result in inadequate disturbance estimates.In adaptive schemes, it is possible to include the voltage variation at different nodes of the system to determine when load shedding should be performed by relating the magnitude of the voltage drop to the amount of load to be shed [4].Due to the massive incorporation of power electronic devices and the consequent loss of inertia in power networks, the implementation of this type of underfrequency load shedding is becoming increasingly complex.
Adaptive UFLS schemes are based on events or responses, frequency prediction, and RoCoF [42].Below are the characteristics of the different adaptive schemes found in the literature, explained in more depth.

1) EVENT-BASED APPROACH
In event-based UFLS schemes, the main consideration for operation results from the capacity of the generator that has gone out of service.This type of UFLS requires a communication scheme to send information or measurements to the control center for operation.This is possible with the development and implementation of a WAMS.Based on the information provided to the control center, the load to be removed is usually calculated using Equation (3) [4].
where P is the power loss, P res is the available reserve that rotates in the system, and P elim is the power or load to be disconnected [43].The main drawback of this technique lies in the need for fast and direct communication between the different generating stations and the control center.In [44], two algorithms are presented, the first as a response-based algorithm and the second as a combination of response-based and event-based algorithms, using voltage and frequency indexes to determine the magnitude of the load shedding.On the other hand, in [45], an event-based UFLS is proposed, simulating the Iranian grid and its main city, Khorasan, connected by two transmission lines.The objective of the proposed UFLS is to provide stability after the system is forced to operate on two independent islands when considering the output of the mentioned lines.
In [42], an event-based UFLS is presented to be employed in an isolated offshore power system that includes renewable generation and aims to avoid service interruption due to cascading trips caused by low-frequency protections of wind and photovoltaic generators.Additionally, the algorithm's operation is studied under N-1 and N-2 contingencies in scenarios of maximum and minimum demand, minimizing the load shed and maximizing the frequency nadir through Particle Swarm Optimization (PSO).

2) RESPONSE-BASED APPROACH
This is the most common adaptive UFLS technique, which works on the synchronous generator oscillation equation and calculates the disturbance magnitude from it [44].To describe the oscillation of synchronous generators, Equation ( 4) can be used: Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
where N is the number of generators, P m i is the mechanical power expressed in per unit, P e i is the electrical power expressed in per unit, P i represents the imbalance between the generated power and the load, H i is the inertia constant in seconds, f i is the frequency in Hz, and f n is the nominal frequency of the system.By aggregating the total generators' oscillation equations of the system with the intention of finding the total imbalance, equation ( 5) [44], [45], [46] is obtained. where: where f c is the frequency at the equivalent inertia center.Since during a disturbance, the frequency may change and oscillate at different rates for each of the machines in the system, it is necessary to use the frequency at the equivalent inertia center, calculated as shown in equation ( 6) [47].Typically, in the response-based approach, the load to be disconnected is estimated using the equation ( 7) [4]: where K CF is known as the safety factor and is traditionally set at 1.05, but this has led to excessive load shedding in some cases, so several authors choose not to use this value and ignore that constant.P spin is the available spinning reserve, and P shed is the amount of load to be shed [4].

3) FREQUENCY MEASUREMENT BASED APPROACH
The frequency measurement-based approach is generally divided into two approaches.The first one is based on obtaining a prediction of the minimum frequency that the system will reach after a disturbance, or stating a prediction of the time remaining for the system to reach a certain frequency nadir.The second one takes as its main input the measurement of RoCoF, and from this value, assumes the shedding of the load.There is not an established or standard criterion for determining the appropriate RoCoF values or thresholds for the implementation of UFLS's.Since the RoCoF is the result of imbalances within the electrical system, accurate and low-latency estimation is necessary to create efficient and fast controls according to the specific characteristics of each electrical system.In [30], an UFLS model is presented that, despite being based on WAMS operation, uses a frequency prediction method to estimate the load shedding time and the magnitude of the disconnected load.In [48], an adaptive UFLS is shown involving the use of WAMS and the problems caused by active power deficits and variations that occur during transients caused by voltage-dependent loads.It also predicts the operating point trajectory in phase space and tests the model in the Slovenian power system.Reference [49] presents an UFLS approach that uses the second derivative of frequency as an initial source of information to predict the frequency trajectory as a consequence of a disturbance in the power system.In [50], an UFLS is presented using PMUs for frequency and RoCoF, this scheme is employed for the detection of large disturbances in the power system.
In a power system, the values experienced by RoCoF after a disturbance depend on the characteristics of the network itself, generally behaving in proportion to the imbalance caused.Therefore, the larger the disturbance, the higher the RoCoF will be.On the other hand, the values reached by RoCoF after the UFLS action are related to the control responses or actions performed and not to the network composition or topology.Therefore, RoCoF is not suitable for establishing or indicating network restoration [4].
Although there are few research studies and articles published on RoCoF-based UFLS [4], [11], [23], [24], [29], [51], [52], some authors have presented their investigations.In the case of [11] it is presented an UFLS that uses the frequency rate of change to dynamically calculate the load shedding value, which can be adapted to the speed regulator, and aims to use the rolling reserve.In [51], the use of PMUs for RoCoF measurement and centralized load shedding estimation is proposed.Different synchrophasor prediction models, RoCoF prediction techniques, and real-time data sampling rates are compared.In [29], a semi-adaptive RoCoF-based UFLS with multiple load shedding stages was developed.The proposed scheme employs a mixed-integer linear optimization model, and both regulator dynamics and load damping are considered in the frequency response.In [24], a multi-stage UFLS designed based on the annual demand curve is presented, and it is activated by both frequency and RoCoF.
The loss of inertia in electrical systems can lead to restrictions on the penetration of renewable energy or limit the decrease in thermal generation due to the need to provide frequency regulation and inertial response services.Therefore, [12] proposes a new restriction to the oscillation equation to contain the RoCoF and frequency nadir using UFLS services.In [13], UFLS, inertia, and constraints on thermal power plants are employed to avoid major load shedding.On the other hand, [14] presents a stochastic unit commitment model with frequency constraints that, for the first time, simultaneously optimizes energy production while considering the provision of synchronized and synthetic inertia.In these cases, the UFLS is considered part of a set of reserves available to the system operator to ensure optimal frequency control in terms of system costs [12].In some cases, when a transmission line upstream of a load shedding substation is opened, low-frequency relays may operate incorrectly, so [53] proposes an UFLS method that monitors negative sequence voltage, frequency, and RoCoF, allowing for faster and more reliable load shedding that avoids undesirable operations.
Several authors have focused their research efforts on the UFLS area, particularly on the use of computational techniques for its operation.For example, in [74], a load shedding reduction method is presented using a multi-objective particle swarm optimization approach to determine the magnitude of the load and disconnection time in an offshore electrical system near Taiwan.The proposed approach optimized the disconnection load and recovered the frequency faster than traditional UFLS.In [75], an adaptive strategy is designed based on the execution of neural networks and the study of the transient stability of the network.Simulation data on transient stability is used to train the algorithm to minimize load shedding in various operational scenarios.The input neurons are selected as the total electrical generation, the demanded load, and the frequency decay rate.From these neurons, the minimum amount of load shedding is determined to maintain frequency stability.On the other hand, [61] develops an optimal load shedding method in which the total generation, total system load, hydraulic generation reserve, and frequency reduction rate are considered as input neurons for the construction of the Artificial Neural Networks (ANNs) that seek to minimize the amount of load shed.The method was tested on the New England electrical system.In [76], an UFLS is shown to operate using a neuro-fuzzy method to determine the amount of load to be disconnected.The method is tested on the IEEE 300-bus test system.In [77], two methods are proposed for optimizing the location of the load to be shed to avoid collapse due to voltage instability.The first method identifies the shedding location and applies an analytical procedure to determine the magnitude of the load, while the second method directly predicts the magnitude of the load shed.Finally, in [78], real-time load flow synchronization is used to select the appropriate load shedding based on the contingency presented, using genetic algorithms to optimize the amount of load shed.
As discussed in [4], the most commonly used computational technique in UFLS is genetic algorithms.Some examples of this can be seen in [79] and [80].In [81], an UFLS is presented that aims to minimize the shed load by prioritizing low-frequency oscillations in all stages of the scheme.This is based on the Grasshopper optimization algorithm, and the research found that this method had a faster performance than particle swarm optimization and genetic algorithms.
Another computational technique that is commonly used in UFLS optimization is mixed integer linear programming.Some examples of research that use this technique include [20], [29], [69], [82], which aim to minimize the shed load while ensuring system stability.Similar to the previous technique, particle swarm optimization has also been extensively studied for optimization problems with nonlinearities and multi-objectives, such as in [47] and [83].Lagrange multipliers are another technique used in UFLS, although they are not widely used.An example of this can be seen in [52], which aims to optimize the location and magnitude of the shed load.
Reports of UFLS can be found that use combinations of optimization algorithms to increase efficiency and reduce disadvantages.In [62], a combination of PSO and bacterial foraging optimization (BF) is proposed to improve search capability and reduce issues such as parameter dependence and premature convergence.On the other hand, [84] proposes an UFLS method based on equations derived from the Optimal Power Flow (OPF) problem, allowing the user to calculate the minimum amount of load to be shed and provide a feasible frequency trajectory in the event of a contingency.Predictions are obtained through a second-order dynamic model.In [85], the operation stages for the 81L relay in an isolated offshore wind turbine power system are determined using PSO based on Takagi-Sugeno (TS).The objective is to minimize load shedding and maximize the frequency nadir by developing two sets of TS fuzzy rules to adjust the inertia weight and learning factors in the proposed PSO and obtain the optimal configuration.In [86], a polynomial regression analysis technique is used to estimate the power imbalance between generation and load in an isolated power distribution system, and MILP optimization is used to estimate the optimal combination of loads to be shed.

D. WAMS-BASED UFLS
In recent years, WAMS have gained popularity in power systems.These systems have been enhanced by the use of PMUs that allow synchronization with GPS, significantly reducing delays and increasing the quality and quantity of 110974 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
data [23].Figure 6 shows a diagram of an UFLS scheme using PMUs and considering dynamic load.
Compared to traditional UFLS schemes that operate based on a predefined frequency band with threshold values and involving load shedding steps, which can lead to situations where shedding is either excessive or insufficient, PMUs provide the possibility of developing schemes that are more realistic, reliable, and optimal [11], [23].
The uncertainties inherent in PMU measurements should be considered in their applications, particularly in control systems.These uncertainties arise as a consequence of the behavior of PMUs when operating under time-varying conditions such as those caused by transient events [23], [87].This is the case for events caused by load connection or disconnection, which ultimately affect the accuracy of PMU estimations [88].
Several authors have used UFLS strategies employing PMUs for measurement and as the main input.In [23], the impact of phasor estimation algorithms on UFLS schemes based on frequency measurement and rate of change is investigated.Two phasor estimation algorithms, one static and one dynamic, are compared, and the suitability of using PMUs for RoCoF measurement is analyzed.In [89], an adaptive centralized UFLS scheme method is proposed using PMUs to obtain voltage and frequency information.The proposed method considers both active and reactive power in the load-shedding process.Additionally, to delve deeper into PMU technology and the use of wide-area measurements in monitoring, protection, and control of power networks, refer to [90].In [30], an UFLS focused on frequency prediction is proposed allowing for load shedding decisions.However, the proposed method had issues when there were high-frequency gradients in the system, so it is recommended to use the algorithm only as a complement to traditional UFLS.

IV. UFLS AND FREQUENCY REGULATION WITH SMART LOAD
A new concept and philosophy of operation for UFLS may emerge with the use of smart loads.They can help preserve system stability while maintaining acceptable safety margins to prevent failures caused by frequency events.Smart loads can modify their consumption based on a control logic that could replace traditional UFLS operation.The potential for smart loads to decrease or delay load shedding has not been extensively studied in low inertia or non-synchronouspenetrated networks [24].
As a result of new electricity generation technologies and the massive inclusion of power electronics interfaces and renewable energy generation, electrical systems are undergoing changes in their energy composition and topology, leading to schemes such as electric microgrids and UFLS based on inverters or power electronics.An example of this is the strategy presented in [91], which proposes a method of estimating the power deficit between generation and load based on changes in frequency and RoCoF, and is applied to an inverter-based microgrid (IBM).
Some authors have related the concept of smart loads with the UFLS schemes; for example, [24] proposes a multi-stage UFLS scheme that is designed based on the annual duration curve and is activated by the frequency of the grid and RoCoF.Additionally, the potential of smart loads to delay or reduce total load shedding in low-inertia networks is being investigated.The proposed UFLS scheme is redesigned in the presence of smart loads and is formulated as a mixed-integer linear programming (MILP) model.On the other hand, [92] proposes an adaptive UFLS scheme that incorporates the impact of load response.In [93], the emergence of a new direction for UFLS techniques is recognized as a result of the construction of wide-area measurement systems, the development of demand response technologies, and the application of smart appliances.A smart UFLS/ULVS method is proposed based on the active participation of smart appliances.
Smart loads are typically non-critical loads that can be connected to the grid via power electronics interfaces, allowing them to change their level of consumption and perform frequency regulation [94], [95].These can be classified into two main groups: static loads (generally voltage-dependent) [94] and rotating loads (generally frequency-dependent) [24], [96].In [93], an example of thermostat-controlled smart loads is presented, which provide frequency regulation to the system and aim to reduce the load shed by UFLS.Equation ( 8) represents frequency-dependent dynamic loads, while equation ( 9) represents voltage-dependent static loads [24].An equivalent power system is depicted in Figure 7.The upper part shows the generation and transmission systems, while the lower part includes smart loads, backup loads for UFLS, and fixed loads [24].
where P represents the dynamic smart load being calculated; P 0 is the reference load at a specific reference frequency f 0 ; f is the frequency at which the load is being measured; f 0 is the reference frequency corresponding to P 0 ; and kpf is a constant exponent that determines how the load value changes with respect to the frequency ratio f /f 0 .
where P represents the dynamic smart load being calculated; P 0 is the reference load at a specific reference voltage v 0 ; v is the voltage at which the load is being measured; v 0 is the reference voltage corresponding to P 0 ; and kpv is a constant exponent that determines how the load value changes with respect to the voltage ratio v/v 0 .
In general, the amount of frequency regulation that a load can provide by modifying active power consumption is related to its degree of dependence on frequency and voltage and is restricted in the time it provides this service by its capacity curve and its electrical and mechanical limitations [24], [94].Thus, smart loads can be used to perform primary frequency regulation by reducing the magnitude of the load shed by the UFLS, decreasing the RoCoF, and improving the frequency nadir [24].
The equation ( 10) represents the amount of reserve in terms of active power that a particular dynamic load can provide to the grid [94].Furthermore, the response of the smart load can be discretized, as shown in equation ( 11) [97].
where Res max m represents the maximum reserve power that the dynamic load can contribute to the grid, P 0−i,m denotes the rated power of the load under normal conditions.The variable f dr m represents the actual frequency of the grid at which the dynamic load operates, while f 0 is the reference frequency or desired operating frequency of the grid.The coefficient kpf − m reflects the sensitivity of the load's power consumption to changes in frequency.Finally, f dr−min m indicates the minimum frequency at which the dynamic load can operate and contribute reserve power; if the actual frequency falls below this value, the load cannot provide any reserve power to the grid.
where P SL n,m represents the change in power output of the smart load at time step n and mode m, reflecting its response to the grid conditions, P SL n−1,m represents the previous power output of the load, capturing its response in the previous time step.The variable t is the time step or interval between measurements, while T s represents the settling time of the smart load, indicating the time needed to achieve a steadystate response.
If the dynamic smart loads are integrated into the UFLS model, the frequency response of the system is altered, as shown in equation ( 12) [24].
where K n represents the frequency response factor at time step n, indicating how the frequency is altered, H eq reflects the equivalent system inertia constant determining the rate of frequency changes, P gov n,c captures the change in power output of conventional generators due to governor action, P c represents the change in power consumption from constant loads, D f n,c accounts for the damping effect of frequency deviations on power consumption, N s s x s,n,c captures the contributions of non-smart dynamic loads to the frequency response, and N SL m=1 P SL n,m represents the contributions of dynamic smart loads to the frequency response, with the ability to actively adjust their power consumption to support the frequency.This equation quantifies how these factors collectively influence the frequency response in the UFLS model when dynamic smart loads are integrated.
Finally, if one wishes to determine the frequency of the system, it would be determined by equation (13).
where f n represents the frequency of the system at time step n, f 0 represents the reference frequency of the system, and f n represents the deviation in frequency from the reference frequency at time step n.
Some loads, commonly referred to as ''smart loads'' in recent times, are equipped with power electronics interfaces that allow them to dynamically adjust their consumption parameters.Such loads can provide responses to variations or imbalances between generation and load and therefore perform some actions for frequency regulation in advance of the UFLS schemes.This decreases the amount of shedding by traditional UFLS and sometimes delivers the necessary power to prevent their operation.By postponing or avoiding the disconnection action of loads by traditional UFLS, electric grids with reduced levels of inertia can be more flexible in the face of different contingencies [24].The power electronics interface is usually considered for rotating loads and sometimes included for static loads, such as electric springs (ES) that are connected in series with a non-critical load to react to frequency variations [95].
It is important to note that the inclusion of power electronics for the control of static loads increases energy losses and system costs in most cases, so [95] proposes the implementation of an aggregator that serves as control for a specific group of loads that have similar characteristics.The power reserve in smart loads depends on the operation times of each of them since they are not usually in 110976 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.continuous operation 24 hours a day.In addition to this, a fully decentralized control of several loads based on local measurement can contribute to frequency regulation without requiring any coordination or communication [95], [98], [99].Reference [96] proposes a power control that implements small capacitors in the DC link to modify the levels of active and reactive power supplied to permanent magnet synchronous and induction motors.On the other hand, [100] introduces optimal load control through multi-objective optimization based on gain adjustment to minimize frequency nadir, time response, steady-state error, total load loss, and the aggregate disutility of controllable loads subject to power balance in the network.

A. ELECTRIC SPRINGS
According to [95], the concept of smart loads was first introduced in 2012 by [101], where the concept of an electric spring was introduced as an analogy to the mechanical spring, demonstrating its ability to provide voltage support, energy storage, and dampen electrical oscillations.The simplified diagram of an electric spring is presented in Figure 8.It can be observed that the load is connected in series with the elements that will carry out the control.This load can operate normally or provide regulation services if needed.
Electrical systems that integrate more power electronics and, therefore, have lower inertia levels often face problems with damping frequency and voltage oscillations.As a result, several authors have proposed techniques to emulate inertia and do the response to imbalances or faults more damped.These techniques include emulating the behavior of synchronous generators called virtual synchronous generation (VSG) [102], [103] using battery energy storage systems (BESS) [104], HVDC VSC links [105], [106], and other proposals that typically focus on managing the supply side and not the demand side [107].
ES are devices that are connected in series with non-essential electrical loads, such as those intended to be part of the UFLS schemes, and that react to network variations to provide damping.Essentially, UFLS schemes are conceived as the last control action to maintain stable levels of active power and protect power systems against frequency collapses [108], so the load that is shed (noncritical load) represents a portion of the system that is left without power supply for a period of time as a consequence of the scheme's action.This makes the operation of UFLS not always desirable.In this case, ES could be used to reduce the action of UFLS in the face of power imbalances affecting frequency levels and thus avoiding unnecessary load shedding.
Electrical springs can be considered an emerging technology that enables the conversion through smart loads and has received much attention due to its ability to perform frequency and voltage regulation actions with a reduced requirement for energy storage capacity [107].Some electrical springs have been employed for voltage regulation and frequency control in AC microgrids [101], [109], [110] and in DC networks [111], [112].
Electrical springs have not only been employed and investigated for use in microgrids and distribution networks but also in large power systems.For example, in [113], a dynamic ES is presented incorporating control design and power electronic circuit dynamics, demonstrating the model with a radial-cordial decomposition controller for voltage and frequency regulation.On the other hand, [114] addresses the problem of the conflict between network frequency regulation and economic power flow.This is achieved with the conventional control scheme caused by the coupling between active and reactive power of smart loads and proposes a centralized control that operates multiple ES based on frequency regulation and optimal energy flow.
Originally, electrical springs were proposed to mitigate intermittent energy generation variations such as solar and wind power by performing demand-side management actions without using communication schemes [115].Conversely, [116] proposes the use of a consensus control for a group of ES through a WiFi communication layer that modifies active and reactive power levels, performing frequency and voltage regulation.In [117], a control scheme for ES aggregators in subtransmission networks is proposed for frequency and voltage control based on the adoption of the d-q transform.Reference [118] proposes a state-space model for an ES embedded in a distributed generation power system, considering supply voltages and transmission line impedances.On the other hand, the strategy proposed in [119] focuses on the frequency stability problem and proposes a control for ES that operates in capacitive or inductive mode depending on frequency fluctuations, with simulated case studies using MATLAB/Simulink.Other papers have studied different aspects of ES, such as its dynamic behavior [120], performance analysis [121], [122], control [123], [124], effectiveness of ES and STATCOM in voltage control [125] and effectiveness of smart loads in frequency control [95], [109].

B. SMART INDUCTION MOTOR DRIVE
Traditionally, rotating loads being directly coupled to the electrical grid have provided inertia by delivering or absorbing energy in kinetic form when imbalances occur between demand and generation.With the implementation of power electronics interfaces between the grid and the load to control their output parameters such as speed and torque, the existing electromagnetic coupling with the grid is lost, causing the dynamic behavior of both parts to operate independently.Although, from the perspective of the overall system inertia, the implementation of power electronics interfaces may seem undesirable, this interface provides loads with a greater degree of controllability, which could reduce their harmful effects on stability and, conversely, improve the behavior of the power system.Figure 9 shows a basic diagram for a variable speed drive (VSD).It includes rectifier and inverter devices, the DC bus, the external grid, the motor, and the mechanical load coupled to it.
An example of a rotating load that uses a power electronics interface for its operation and connection to the grid is the induction motor, which uses a VSD to modify its output characteristics.Using loads in frequency regulation offers some advantages, such as faster response times, reduced fuel consumption, emissions, and better disturbance location [126].VSDs are particularly suitable for frequency regulation because their demand changes with changes in speed while most loads can only be on or off [127].The development of VSDs, their massive implementation, and the increase in energy costs may open a door to the provision of ancillary services from the load side [128], [129].
Some authors have devoted their work to the development of strategies for using VSDs to provide frequency regulation to the electrical grid.For example, [96] proposes the use of small capacitors in the DC link of three-phase inverters to control the power taken from the grid and compensate for the imbalance between generation and demand.Reference [130] proposes the use of a frequency inverter connected to a motor to perform primary frequency regulation and emulates the inertial behavior of a motor connected directly to the grid.The dynamic limitations of the drive are considered through a speed limiter.The impact of the speed rate, maximum reserve power, and inertia of the load coupled to the motor on the dynamic behavior of the VSD during the process of supporting frequency on the electrical grid is studied.Reference [131] develops a closed-loop control for VSD to provide primary frequency support and develops a real-time energy emulator for a passive-front-end VSD that uses a three-phase Voltage Source Converter (VSC).

C. THERMOSTATICALLY CONTROL LOAD FOR FREQUENCY REGULATION
In order for a controlled load to help regulate frequency in an electrical system, it should vary its energy demand without affecting the process and comfort of the end user [132].Some authors point out Thermostatically Controlled Loads (TCLs) as promising for providing frequency regulation services [133], [134], [135].References [136], [137], and [138] describe some research works that investigate the operation of TCL based on a specific power profile.The control of TCLs typically involves three methods: direct on/off switching, temperature set-point variation, and a hybrid control that combines both on/off switching and temperature set-point variation [132].

1) DIRECT ON/OFF SWITCHING TCL
Direct on/off switching TCLs are typically devices that, upon receiving an operational instruction (such as a defined frequency threshold), switch their state to provide regulation services to the grid.Some of these TCLs can be operated through aggregators to provide primary frequency regulation and, in some cases, deliver virtual inertia to the power system [132].
An aggregation of TCLs that provide frequency regulation services to the grid must consider both the end-user temperature and the grid frequency.Figure 10 shows a way in which these parameters could be related.A dead band is considered in which the aggregation does not change state, a range of values in which the device turns off (to increase frequency or temperature), and another in which the device turns on (to decrease frequency or temperature).Nominal operating thresholds are also included.
TCLs have been considered to follow an arbitrary power profile derived from frequency measurements, in the case of [139] where the aggregated flexibility offered by a collection of TCLs is modeled as a stochastic battery with dissipation.In [140] the potential of electric domestic hot water heaters for demand-side management via pseudo-cost functions is studied.The optimization problem is formulated as a binary integer program using a fully mixed thermal model of the water heater, while actual system behavior is simulated using a multi-layer model.The paper [141] presents the results of an implementation of autonomous optimization for demand-side management of domestic hot water heaters.The optimization is based on one-way communication of pseudocost functions.
Reference [134] explores the feasibility of utilizing clusters of residential TCLs, including air conditioners, to engage in intraday wholesale electricity market price arbitrage while utilizing non-disruptive load control methods.The research presents two approaches for arbitrage: the first is a benchmark that provides an optimal policy but requires either local computing or real-time communication.The second approach, which relies on a thermal energy storage model, requires less computation and communication infrastructure but is not as optimal as the benchmark method.The paper [142] outlines the design considerations for a centralized load controller responsible for regulating thermostatically controlled appliances (TCAs) to provide continuous regulation reserves (CRRs).The logic of the controller is based on establishing baseline loads, creating priority lists, sending dispatch commands, and fine-tuning the simplified forecaster model obtained from measurement data.

2) TEMPERATURE SET-POINT VARIATION TCL
Temperature set-point variation TCL corresponds to thermostatically controlled loads that, in response to network variations or operational instructions (such as frequency or RoCoF), change their temperature set-point.By changing the temperature set-point, the energy demand can be varied and used to help improve the imbalance in the grid.
Reference [143] develops a mathematical model aimed at the design of a model-based feedback control strategy, air conditioners are used and are divided into groups according to the precision of their thermostats in order to improve frequency response.Paper [144] presents a framework for controlling network frequency by simultaneously involving resources on both the generation and demand sides through a fast, transactive control approach.Initially, a proportional frequency-price relationship is used to build and analyze a transactive frequency droop controller for a single-area power system.Subsequently, a transactive demand response system is developed by incorporating a large population of thermostatically controlled air conditioning loads.
A proportional-integral controller is used to adjust the set temperature of the air conditioners based on price variations.The Network Code on Demand Connection proposed by ENTSO-E includes a mandatory frequency support service for thermal loads.However, in [145], this implementation approach leads to an uncertain response that is highly dependent on the specifics of the controller used.The authors present a case study that showcases significant variations in frequency response patterns among three controllers, all of which are compatible with the proposed Network Code.
3) DIRECT ON/OFF SWITCHING AND TEMPERATURE SET-POINT VARIATION TCL Some authors have worked on developing control logics that include on/off thermal cooling loads (TCLs) and temperature set-point variation TCLs.
Reference [146] proposes a stochastic control of TCLs that enables modulation of the aggregate energy consumption of a large collection of devices.The paper [147] shows that TCLs (domestic refrigerators) can be controlled without real-time communication and in a nondisruptive way to collectively enhance the network frequency response.The aggregated power consumption of TCLs, distributed across the system, could be controlled as a 'linear' function of the locally measured frequency and its rate of change.Alternatively, their aggregated consumption could be made to follow a 'preset' power profile depending on the estimated infeed loss.A novel technique for accurate estimation of infeed loss and consequent postfault TCL power reduction is also proposed.

V. ADVANTAGES AND DISADVANTAGES OF UFLS TECHNIQUES
In this section, the main UFLS techniques are compared based on their advantages and disadvantages.Traditional UFLS techniques are the simplest and therefore the easiest to implement.However, they often provide an efficient but not optimal response to frequency imbalances, which can lead to shedding excessive or insufficient load.Additionally, their configuration is always static, and it is not possible to make parameter changes online.On the other hand, adaptive UFLS schemes based on event, response, and frequency measurement can adapt to network conditions and shed a load portion closer to the required level.Frequency prediction-based techniques have the ability to shed load fairly accurately, while techniques based on RoCoF may experience variations due to their sensitivity to network oscillations.Table 1 provides a classification of UFLS techniques and compares their advantages and disadvantages.

VI. CONCLUSION AND FUTURE
Multiple UFLS schemes have been proposed to regulate the frequency of the electrical system and prevent blackouts.Although reviews have been conducted on these schemes, smart loads have not been considered important elements that could contribute to their design and performance.The main objective of electrical systems is to safely and reliably supply power to all loads.Smart loads create a new pathway for designing new UFLS schemes that reduce the amount of disconnected load while maintaining the stability and security of the electrical system.UFLS schemes based on smart loads are a new and innovative proposal that has not been thoroughly studied so far and could be a promising area of research that provides tools for the electrical systems of the future.
The main contribution of this paper lies in its comprehensive examination of UFLS schemes, with a specific focus on the integration of smart loads and their impact on frequency regulation.By conducting a literature review encompassing existing UFLS schemes, including those utilizing smart loads, the study sheds light on a previously overlooked aspect of UFLS design and performance.The findings highlight the potential of smart loads as essential elements in developing innovative UFLS schemes to minimize load disconnections while ensuring system stability and security.This exploration opens up a promising area of research that can significantly influence the concept and philosophy of UFLS schemes, paving the way for the electrical systems of the future.
Moving forward, several future directions can be pursued in UFLS studies to build upon the research presented.First, in-depth investigations are needed to determine optimal integration and control strategies for smart loads in UFLS schemes, considering the varied characteristics and response capabilities of different load types.Additionally, the economic and environmental implications of UFLS schemes incorporating smart loads, such as the potential for demand response programs, should be further examined.Furthermore, advanced communication and coordination techniques between smart loads, distributed energy resources, and grid operators warrant exploration to enhance the effectiveness of UFLS schemes.Continued research in these areas, coupled with an assessment of reliability, resilience, and regulatory frameworks, will facilitate the development and implementation of efficient UFLS schemes based on smart loads, propelling the evolution of electrical systems.
The future scope of this review will enable advancements in the integration and development of novel UFLS schemes through the implementation of smart loads.This holds the potential to reshape the operational philosophy of UFLS schemes, significantly mitigating load disconnections while simultaneously enhancing frequency response in the electric system.This is particularly significant within the context of emerging electrical systems, which are oriented towards a general reduction in inertia.

FIGURE 2 .
FIGURE 2. Characteristic operating time for frequency regulation.

FIGURE 3 .
FIGURE 3. General flowchart of the review work.

FIGURE 4 .
FIGURE 4. Traditional load-shedding scheme according to ENTSO-E at a nominal frequency of 50 Hz.

FIGURE 6 .
FIGURE 6. Diagram of a UFLS scheme using PMU.

FIGURE 7 .
FIGURE 7. Smart loads and their operation to provide frequency regulation.

FIGURE 8 .
FIGURE 8. Simplified diagram of an electrical spring in series with a non-critical load.

FIGURE 9 .
FIGURE 9. Simplified diagram of an smart induction motor drive.

FIGURE 10 .
FIGURE 10.Relationship between temperature, frequency, and TCL aggregation for grid frequency regulation.