Forced Oscillation in Power Systems With Converter Controlled-Based Resources— A Survey With Case Studies

In future power systems, conventional synchronous generators will be replaced by converter controlled-based generations (CCGs), i.e., wind and solar generations, and battery energy storage systems. Thus, the paradigm shift in power systems will lead to the inferior system strength and inertia scarcity. Therefore, the problems of forced oscillation (FO) will emerge with new features of the CCGs. The state-of-the-art review in this paper emphasizes previous strategies for FO detection, source identiﬁcation, and mitigation. Moreover, the effect of FO is investigated in a power system with CCGs. In its conclusion, this paper also highlights important ﬁndings and provides suggestions for subsequent research in this important topic of future power systems.


I. INTRODUCTION
In power systems, forced disturbances (FDs) such as unexpected equipment failures, uncertain inputs in turbines, control interactions, and abnormal operating conditions are relatively common [1], [2]. A new oscillation behaviour has been observed in power systems. It has been shown that this new oscillation phenomenon cannot be reproduced by traditional models and methods applied in power systems [3]. The FDs with different ranges of frequency may provoke forced oscillations (FOs) [4]- [6]. If the frequencies of the FDs coincide with the frequencies of electromechanical (EM) modes, i.e., normally in the range of 0.1 Hz to 2.0 Hz, this event will lead to a resonance resulting in a severe FO [4]- [6]. In practical power systems, FOs occur in a number of places in USA, Canada, and China. They normally persist in the system for minutes or even hours. These scenarios may degrade the power system stability, power transfer capability, and they may damage the relevant devices [1], [2], [4]- [6].
The associate editor coordinating the review of this manuscript and approving it for publication was Siqi Bu .
Several techniques have been applied successfully to mitigate FOs. In control and power electronic systems, the voltage source converters connected to long transmission lines may be cause of FOs due to the resonance from a passive property of conductance. A frequency-domain passivitybased voltage source converter control (VSC) [7], and a virtual-impedance-based control for VSC and current-source converter [8], can be used to deal with this problem. Alternatively, this issue can be resolved by shaping both filter impedance and closed-loop control output impedance, and by providing damping in specific frequencies. A review on damping of sub-synchronous torsional interactions by using VSC-based flexible AC transmission systems (FACTSs) and damping controllers can be found in [9]. These challenges will be emerging into the future electronics-based power systems [8]. In power systems, the most common methods to mitigate FOs are: cancelling the FO sources [10], keeping the FD frequency far away from the electromechanical oscillation (EMO) frequency [10], alleviating the FD magnitude [10], and increasing the damping ratio of critical modes by the following methods: i) using power system stabilizers (PSSs) in synchronous generators (SGs) [11]- [13], ii) power oscillation dampers (PODs) in renewable energy sources (RESs) and FACTS devices [14]- [16], and iii) active/reactive power modulations via converters of converter controlled-based generations (CCGs) [17]).
Future power systems will contain a substantial number of CCGs. Consequently, power systems have been transformed significantly over the last few decades. Two new standards have been introduced in terms of power system dynamics and stability, i.e., converter driven stability and resonance stability [18]. Therefore, the CCGs, i.e., renewable energy sources and battery energy storage systems, are expected to play an important role in future power systems [19]. The penetration of CCGs may result in FOs due to the stochastic characteristics associated with them. As a result, the penetration of CCGs in future power systems can contribute to FOs and severe interaction with system modes. For instance, an FO could be caused by the sub-synchronous interaction between voltage source converters and the grids [20]. Previously, the major challenge of substituting SGs with CCGs was not well studied or fully understood [21]. Moreover, the CCGs may create additional oscillation modes associated with their converter controls [22]. Frequency of the new oscillation mode (or FO mode, which is later introduced to the system) may coincide with the EMO frequency resulting in interaction among adjacent EM modes, and a greater complexity to analyze characteristics of FOs and EMOs in power system with CCGs.
The goals of this paper are to consolidate prior works, research consensus on the FO, and provide the researchers a broader picture by: 1) Detailing general backgrounds of the FOs in power systems with and without CCGs, and indicating similarities and distinctions of the FOs, EMOs, and limit cycles; 2) Presenting holistic reviews of actual FO events, as well as FO detection, identification, and mitigation methods in traditional power systems in order to find research gaps, and consequently suggest possible challenges and issues for FOs in future power systems with CCGs; 3) Conducting simulations of various case studies in a future power test system to analyze the effects of FOs caused by CCGs.

A. FUNDAMENTAL OF FO IN POWER SYSTEM
Power system oscillation can be divided into three major categories, i.e., oscillations from ambient, transient, and forced input as demonstrated in Fig. 1 [23]. Three major types of oscillation are briefly described as follows [24]: i) Ambient is the system response to minor changes. This change is typically characterized by white noise, small random load fluctuations, and stochastic outputs from CCGs. The ambient always occurs in the data obtained by measurements, and it can be handled by using filters; ii) Transient is the response subjected to a change in the power system from an equilibrium or a steady-state operating point such as a fault, outage contingency, line and generator trips, or load rejection. Oscillations in a transient response are naturally characterized by the critical EMO or system modes; iii) Forced is the response in systems associated with an external input, a malfunctioning apparatus, or uncertain inputs such as a malfunction in the on/off cycling of steam valves, furnace induced dynamics, a rapid change in vortex, a malfunction of converter control loop, or poorly designed controllers [3], [25]- [28]. Forced oscillations may include harmonics of the fundamental frequency of the forcing function. Therefore, it may result from the periodicity of the external inputs. The FO is typically undamped and sustained. Moreover, the FO may occur until the FD disappears, or the FD is removed from the system. Specifically, the FO tends to be more severe when frequencies of the FD coincide with critical electromechanical (EM) modes [4], [11], [12].

B. COMPARISON TO EMO
In comparison, the FO is quite distinct from the EMO. The well-known EMO is caused by the dynamic interaction among the groups of synchronous generators, while the FO is excited by an exogenous periodic perturbation with various frequency bands. It means that the form of an EMO or natural response depends on the system operating points, while the form of an FO depends on the driving FD input. This distinction explains why natural responses can be analyzed to estimate the system inter-area and other EM modes [3], [29]. The FO can occur in power systems under a wide variety of circumstances such as failure in equipment, poorly tuned controllers, and abnormal generator operating points [24]. The EMO can be analyzed by the small-signal stability assessment using exact or estimated system parameters and models. The EMO can be suppressed by improving the damping ratios of EM modes. However, the sustained FO can occur after exciting by external perturbations at frequencies close or equal to those of EM modes [5]. The FO is spurious because it is later introduced to the system as external perturbation. Therefore, the FO cannot increase the order of the system or create additional state variables. The FO exhibits an extremely high amplitude due to the resonance effect and it may lead to wide-area blackout, particularly in weak power systems [30]. During that period, the response of the FO comprises two components, i.e., EMO and FO. These may consequently disappear after the ending time of the FO [4]. Normally, the damping of the FO component is around zero, and the frequency of the FO component is the same as that of FD [5]. However, the EM mode may interact with the FO. This scenario may consequently result in alterations to damping and the frequency of the EM mode [11]. Additionally, the mode shapes of the FO and EMO are almost identical. Therefore, conventional EMO detection methods cannot be applied to extract the EMO in the presence of the FO [11].

C. COMPARISON TO LIMIT CYCLE
Limit cycles due to the supercritical Hopf bifurcation are another reason for major oscillations in power systems. The limit cycles are nonlinear phenomena occurring in power systems [31], [32]. The instance of limit cycles indicates the loss of power system stability. To restore stability, some operating points may need to be reconstituted [33]. The oscillation behaviours of limit cycles and FOs are identical in time series data. Both of them can result in an unstable system, leading to a difficulty to locate exact causes and apply effective control actions [33]. However, major distinctions between the FOs and limit cycles can be listed as follows: 1) The limit cycles are inherently nonlinear phenomena that can only be described by nonlinear equations while the FOs are excited by external disturbances [33].
2) The characteristic signatures of the power spectrum and auto-covariance of the FOs and limit cycles can be differentiated by analyzing random perturbations [34].
3) The oscillations resulted by limit cycles do not depend on initial conditions. On the other hand, the oscillations resulted by FOs are related to initial conditions and forcing functions [35]. 4) To avoid the limit cycles, countermeasures such as generator dispatch, load tap changer blocking, and emergency oscillation controls, are suggested [34]. For the FOs, the countermeasures against forced oscillation are to locate and remove the external oscillation sources [33]. Besides, the FOs can be suppressed by the means of power oscillation damping [11], [12], or modulation from power electronic converters [13], [15].

D. FO IN POWER SYSTEMS WITH CCG
In future power systems, synchronous generators will be replaced by CCGs. Hence, this scenario will lead to system strength and inertia scarcity. Therefore, the problems of FO will emerge with new features of the CCGs. Table 1 summarizes the comparison of characteristics and identification between the FO in power systems with and without CCGs. With the increasing penetration level of CCGs, dynamics of future power systems are more unpredictable due to uncertain CCGs. According to Table 1, scenarios of the FOs will be intensified by CCGs, leading to difficulties in the investigation of oscillation, frequency, and damping characteristics.

E. MODELING OF POWER SYSTEMS WITH FO
A general form of a power system model under the occurrence of the FO can be expressed by a nonlinear equation asẊ(t) = f x X(t), U(t), d(t f ) , where f x is the nonlinear function of state variables (X ), t is the moving window time, t f ∈ t is the FO time, dX dt is the differential of X with respect to t, Y is the output vector, U is the input vector, and d is the forced disturbance. To evaluate system stability, the nonlinear model is linearized by (1a) and (1b) [11], [12], A is the state matrix, C is the output matrix, D is the feedforward matrix, B c is the input matrix associated with control vector U , and B f is the input matrix where d is injected.
In power systems with CCGs, X(t) and B f can be obtained in the form of conventional SGs and CCGs as, where X SG and X CCG respectively represent the state variables of all SGs and CCGs, and B f ,SG and B f ,CCG respectively represent the input matrices associated with the FD of SGs and CCGs.
In (2), the state variables X CCG are used to represent the additional dynamics of CCGs, and these variables significantly affect the power system dynamics and stability. In (3), the input matrices B f ,CCG mean the locations of FDs in CCGs such as wind model, solar irradiation, PI controller, high-level control loop, inner-current control loop, and phase locked- where d = d i , · · · , d n d , i = 1, · · · , n d is the counter of FD, and n d is the total number of FD (for example, a single FD is represented by i = 1 and multiple FDs are expressed by i > 1). In (4a) − (4c), it is acceptable that the identification of a single FO source is easier than that of multiple FO sources. Specially, uncertainties from CCGs may result in greater complexities to analyze multiple FO sources. For multiple FO sources, the ranking of FO sources may be applied (please see Section IV-A for a review of FO source identification).
A combination of several sine waves is usually used to model the FD since it can create a severe FO [37]. Other FD models such as square waves, impulses, and random noises can also be considered as the FDs (if they could trigger discernible oscillations with specific frequencies). The FD models in terms of mathematical equations are presented in [11], [12], [37]. For example, Fig. 2 shows the FD models. Fig. 2(a) is the pure sine wave analyzed by [12], while Fig. 2(b) is the combination of sine waves as reported in [11], [12]. Multiple FDs are also considered in [11]. The periodic FD in Fig. 2(c) is investigated and modeled by [37]. Fig. 2(d) is the non-stationary FD. It is believed that this type of FD leads to the non-stationary FO [38]. However, the impacts of the non-stationary FD on the power system stability are yet to be fully analyzed and understood.
Several practical causes of the FD models in actual power systems are given as follows: 1) In 2012, the Midcontinent Independent System Operator (known as MISO) performed base-lining analysis to set inter-area oscillation monitoring thresholds. However, the base-lining analysis identified an instance of the FO with 0.285 Hz for two minutes and a half [28]. In this event, the cause of FO was malfunctioning of the turbine controller after a routine valve testing. An investigation by the plant's personnel confirmed that the FO was driven by a single FD, such as Fig. 2 (a), with frequency around 0.285 Hz; 2) The Electric Reliability Council of Texas (ERCOT) observed several examples of oscillations driven by controllers of wind generators under high wind output conditions. The ERCOT and its consultants conducted pre-and post-disturbance spectral analysis of the PMU data following significant wind events on 3 November 2010. They found that several FO frequencies consisting of 3.2, 5.0, 5.4, and 5.5 Hz, were prominent in wind generation areas [28]. Accordingly, the wind generators could create a FD with a combination of multiple frequencies, as depicted in Fig. 2

(b); 3) In November 2016, a periodic FO was detected in the
American Electric Power (known as AEP) footprint from a PMU [28]. This PMU was used to monitor a solar power plant. The FO was caused by a periodic solar irradiation during daytime. In this case, the periodic solar irradiation can be assumed to be a forced disturbance, as in Fig. 2 (c); 4) An FO driven by a non-stationary FD in Fig. 2 (d) probably occurred on 3 October 2017 in the ISO New England (ISO-NE) power system when multiple frequencies with growing magnitudes were caused by a large generator located outside of the ISO-NE power system [39]. In this event, the fundamental of FD frequency was linearly increased from 0.1 Hz to 0.3 Hz in 100 s, covering the inter-area oscillation frequency bands. The obtained results indicated that the nonstationary FO can significantly excite all the system modes. When the mentioned FDs are injected into the system, the closed-loop state equation can be obtained by the following formẊ cl (t, t f ) = A cl (t f )X cl (t f ), where X cl and A cl are the closed-loop of X and A including d, respectively. The measured outputs with FO (Y cl ) can be written by a closed-loop system as in (5a) and (5b), where e is the exponential constant, C cl is the closed-loop state matrix, emo = 1, · · · , N emo is the number of EM modes, F emo is the frequency of corresponding emo th modes, N emo is the total number of EM modes, S emo and S d are the amplitudes of corresponding emo th EMO and d th FO modes, α emo and α d are the coefficients of corresponding emo th EMO and d th FO modes, and ζ emo and ζ emo are the damping coefficients of corresponding emo th EMO and d th FO modes. During the FO, the signal Y cl contains both EMO and FO. This signal is measured by a phasor measurement unit (PMU) with a time stamp in the range of 40 to 100 ms [40]. The signal Y cl can be used to analyze the FO components, i.e., F emo , ζ emo , F d , and ζ d . Let us consider (5b): the FD creates twin components, i.e., EMO (first term) and FO (second term). The first term is dominated by ζ emo . The higher the value of ζ emo , the lower the value of α emo S emo . Similarly, the second term is dominated by ζ d . The higher the value of ζ d , the lower the value of α d S d . After getting the A cl , stability indices, i.e., the eigenvalues (λ cl ) and damping ratio (ζ cl ), can be estimated. Let λ emo and λ d be the eigenvalues of the EMO and FO modes, when {λ emo , λ d } ∈ λ cl , frequency of EMO (F emo ) and FO (F d ) modes in Hz can be obtained by: As reported in [4], a smaller F means a higher interaction. However, when F = 0 (i.e., F emo = F d ), strong resonance between the EMO and FO modes may occur [4]. This could lead to the maximum value of α emo S emo and α d S d [4], [5], [11], [12]. Both cases may jeopardize the oscillatory stability of the power system. In the latter case, the FO and EM modes are merged. Under this scenario, it is difficult to distinguish the EMO from FO modes [6], [41], [42]. The damping and frequency of such a scenario can be estimated. These estimated modes are referred to as resonance modes [6], [11], [12], [41]. The severity of the FO also depends on the FD model, amplitude, and phase [4], [5], [11], [12]. These analyses can be further used for making control decisions to mitigate the FO [11], [12].

F. MAJOR FO EVENTS IN POWER SYSTEMS
There have been 14 major FO events recorded since 2005 by the North American Electric Reliability Corporation (known as NAERC) [24] and the North American Synchrophasor Initiative (known as NASPI) [28]. The FOs in these events were measured and analyzed by data from PMU or supervisory control and data acquisition (SCADA). To observe the oscillation in power systems, it was reported that PMU is more suitable than SCADA due to a lower time stamp [3], [28]. The FOs in these events were analyzed by Oscillation Detection Module, Mode Meter Module, Oscillation Monitoring System (including Event Analysis Offline and Damping Monitor Offline), and Pattern-mining Algorithm. As can be summarized in Fig. 3, significant observations of these events are highlighted as follows: 1) The FO frequency could vary between 0.01 and 14 Hz, and the FO duration could be in the range of seconds to several hours. However, the oscillation magnitude is higher in a lower frequency range (approximately 100 − 200 MW); 2) Active power, reactive powers, voltage, and current can be used to observe and analyze the FO phenomenon; 3) Most of these FO events were resolved by the network solutions, i.e., reducing the FO source output, and repairing and tripping the FO source [24], [28]. Some events were resolved by the control actions, i.e., switching of control mode and re-tuning of control parameters; 4) Suitable control actions were not applied for at least 11 events due to limitations of measurement, and the lack of fast FO analysis, detection, and source identification algorithms [24], [28]. Besides, these analyses (FO analysis, detection, and identification algorithms) were conducted offline; 5) Most of the FOs occurred due to the malfunction and/or interaction of among controllers, especially the FOs caused by CCGs, i.e., wind and solar generations. As can be observed, the resonance with the EM modes can be caused by the FOs at various locations in power systems. Besides, suitable algorithms for FO analysis, detection, and source identification are highly required; 6) In actual power systems, the steps for mitigating the FOs are implemented by the following steps: implementation of the FO analysis and detection, identification of the location of the FO source, and elimination of the FO by using the network solutions. Nevertheless, the network solutions require system reconfiguration. Thus, the presented methods may further affect power system stability. Therefore, investigations of the FO phenomenon with respect to the power system stability are sought.

III. ANALYSIS AND DETECTION
Significant works have been conducted on this after the major event on 29 November 2005 (see the first event in Fig. 3).
In Western interconnection, a 20 MW FO with 0.25 Hz  [30]. It was concluded that the SSI provided a better estimation result than that of the FFDD. In addition, the SSI can simultaneously estimate the EMO and FO modes (also demonstrated by the same authors in [4]). The performance of the stochastic SSI was also evaluated in [43].
Coherence-based techniques such as the state-space model and AutoRegressive Moving Average eXogenous model, for distinguishing FOs from natural oscillations have been reported in [29]. These methods can be applied to the signals obtained from measurements [29]. However, FOs observed in various frequency bands are difficult to identify using such methods. The detection of FOs in power systems with multi-channel methods was proposed by [44]. In this research project, the performance of FO detection algorithms can be improved by simultaneously analyzing data from multiple PMUs. As a result, a high reliability in detecting small oscillations and faster detection of large oscillations can be achieved. However, this method contains some drawbacks, e.g., its dependence on estimates of the ambient noise spectra, and generalization of the Magnitude-Squared Coherence spectrum. To focus on an interaction of the FO and EM modes, the same attempt was conducted in the two-area four-machine interconnected power system (Kundur) [4]. The result demonstrated that a 10-MW FO can eventually lead to 477 MW of oscillation amplitude in the tie-lines. In addition, an analysis of sensitivities to different FO locations and amplitudes was conducted. The results showed that the FO effects on local modes were relatively small as compared to those of inter-area modes. In [5], the componentanalysis-based FO was mathematically investigated. The words ''beat'' and ''resonance'' were described. An envelope of oscillation signal was also applied to detect the FO. It was demonstrated that the component-analysis-based FO detection method showed a better capability to detect the FO with multiple FDs, and the mixture of FOs and EMs compared to the envelope shape method. The proposed method in [5] was also verified in the Shandong power system (China) on 18 June 2012. Switching the digital electro-hydraulic governing system was utilized to handle the FO. By using the component properties and envelope shapes, five components (i.e., 0.9824, 0.9984, 1.3184, 1.4356, and 2.4255 Hz) were also detected from the measured active power flow in the tieline. Twin components from the FO created the resonance of 0.9824 and 0.9984 Hz. However, the FO source was obscured due to the large time stamp of measurement (around 0.2 s or 200 ms), and the non-occurrence of an oscillation event during the field tests. The simultaneous estimation of the EMO and FO was presented in [6]. In this work, an Autoregressive Moving Average (ARMA) model was applied with FD input signals, resulting in the ARMA+S. Additionally, a two-stage least-squares algorithm from the ARMA was derived to incorporate the FO. Thus, this enables the proposed algorithm to estimate characteristics of the EMOs and FOs simultaneously. The EMO estimation under the periodic FO was proposed in [37]. It was reported that the estimated EM modes can be altered and rely on characteristics of the FDs. The modified Yule-Walker method with estimated autoregressive coefficients of an ARMA was applied to examine the EM modes under the periodic FOs. Moreover, the effects of miscalculation of FO frequency on the ARMA+S was reported in [45]. In this work, the FO frequency estimated by ARMA+S was unable to specify acceptable accuracy in certain cases. The authors suggested an adaptive frequency estimation method to overcome this problem. However, the simulation environment in this study assumed that the EM mode was not changed during the FO. Moreover, it was supposed that the frequency and amplitude of the FO were constant. However, as reported by [38], this may not be a valid assumption since the FO is non-stationary in actual power systems. Therefore, the EM modes may be changed by the system operating points. Consequently, the FO frequency and amplitude may vary according to moving window time. The periodic FO detection was proposed by [41]. In [41], the proposed method used a threshold that is varied by dominant frequency. This was employed to characterize the nature of Synchrophasor measurement unit (SMU). Analysis data were reported in multiple segments to improve the detection performance in the real-time environment. Although the FO components were too small, the proposed method regarding the probability of detection motivated the use of multiple detection segments. The periodic FO detection was enhanced by incorporating a multi-taper approach [42]. In [42], the Thomson's multi-taper spectral estimation and harmonic analysis technique were proposed. This approach is a measurement-based method that requires only the measured data from SMUs for 10 s. By testing on simulated and real SMU data, the proposed algorithms in [41], [42] were applicable for the periodic FOs in practical power systems. In [46], the interaction of the FO with other system modes was analyzed. The forecasting-residual spectrum analysis was applied to discriminate between FOs and natural oscillations [47]. In [48], the wavelet ridge technique was used to track the oscillation patterns and classify modal or FO modes by using different characteristics of noises. A contrastive analysis was applied to differentiate general and special FOs [49]. All of these methods have low-computational demands for utility application. With the small computational time, these methods are claimed to be suitable for reporting the FO analysis and detection results. However, the method proposed in [46]- [49] are suitable only for offline FO analysis. Therefore, such methods may not be suitable to measure the interaction of the FO and EM modes in moving window or real time. In [38], the non-stationary FO was analyzed by using Fourier synchrosqueezing transform. This FO can be caused by the FD containing multiple frequency bands (see Fig. 2 (b)), which also varies in time domain. The dissipating energy flow method was applied to extract the oscillation components and locate sources of the non-stationary FOs. As reported, this type of FO is more realistic in actual power systems. However, the interaction of the non-stationary FO and EM modes was not well reported.

IV. IDENTIFICATION AND MITIGATION A. IDENTIFICATION OF FO SOURCE
As reported in Fig. 3, the FO mitigation methods were mostly conducted by using network solutions. Therefore, the exact locations of FO sources need to be found in order to apply effective actions. Various approaches as reported in [33], [50]- [66], were presented to examine and locate sources of the FO. Table 2 summarizes the methods for the source identification. The major conclusions of this review can be drawn as follows: 1) Most of the identification methods were used to detect FO sources in SG and load. Only two works have proposed the methods for locating the FO in a highvoltage direct current and doubly-fed induction generator, respectively. Thus, locating the FO caused by CCGs in future power systems is sought; 2) To mitigate the FO, an effective way is to locate the source identification before disconnecting it from service. Hence, the FO source identification is a prerequisite for making control decisions. Failing to identify the FO source may result in a false alarm. Thus, the computational time and accuracy of the identification should be small and accurate so that the control action can be performed on time. Nevertheless, locating the FO source remains a challenging task since the FO is sporadic in nature, and it is difficult to be predicted [51], [52]; 3) As proposed by [62], the online management system is quite possible for the detection, source identification, and mitigation of the FO by using PMU data. It also provides an actionable information. However, the computational time (or processing time) for each process should be taken into account so that the control action can be appropriately conducted.

B. FO MITIGATION
In addition to the network solutions summarized in Section II-F, a review of FO mitigation methods by control solutions is given in this section. The control solution can be used to modulate controllable devices (i.e., CCGs, SG, and FACTS) to cancel the effects of FOs without locating the FO source automatically [13]. Table 3 summarizes recent control solutions for FO mitigation. The salient observations can be highlighted as follows: 1) Only seven works so far have focused on this area, when the first attempt for FO mitigation using a control solution was established in 2017 [16]. It is evident that most of the stabilizing devices, i.e., PSS, and controllers in CCGs, HVDC, STATCOM, and UPFC, can be used to suppress the FO. However, these studies were conducted in a simulation platform. Besides, FO mitigation in future power systems with high penetration of CCGs is only assessed in [17]; 2) Only [11], [12] considered the damping of FO mode.
Early investigators were not sure about the damping benefits. Therefore, the damping control design for the FO will be introduced later; 3) In actual power systems, the FO source is durable.
Although the control solution can mitigate the effect of FO (up to 90% as reported by [13]) without conducting any source identification method, the sustained FO may appear in the system. To manage the FO effectively, the FO mitigation framework with control solutions can be incorporated with the network solutions as discussed in Fig. 3. To avoid propagation of the FO, the effect of the FO can be initially mitigated by the control  solutions. Therefore, the source identification can be implemented to find the FO sources. Consequently, the network solutions can be decided for further FO mitigation. Fig. 4 shows the future 14-machine SE-A power system with CCGs. This system has been modified from an original model that has been reported in [36]. 1 The modification and development of this system are conducted based on the Australian Energy Market Operator's Inertia Requirements, Shortfalls Report with Integrated System Plan [67], and the report from Clean Energy Australia [68]. The system in Fig. 4 has been developed and modified by using MATLAB & Simulink version 2018a. Accordingly, more than 24% of electricity is expected to be generated by the CCGs (i.e., wind and solar generations) at the end of 2020 [68]. With the integration of CCGs, the future Australian power system may suffer from the system strength and inertia scarcity. The WG is represented by the doubly-fed induction generator (DFIG)

V. FO ANALYSIS WITH CCGs
with the sixth-order model. The dynamics of DFIG comprise the rotor speed, d-q axis currents of grid and rotor side converters, flux linkage of rotor, and the DC link voltage. The solar photovoltaic (SPV) is modeled by the sixth-order model including the dynamics associated with the current of photovoltaic arrays, voltage of SPV, induce current of DC-DC converter, d-q axis currents of DC-AC converter, and DC link voltage. In this system, there are three dominant modes with frequencies around 0.215, 0.385, and 0.442 Hz, respectively. The damping ratios of these modes are less than 3%. To monitor the FO, the mean frequency deviation ( F ) is measured because of high observability. Moreover, F contains all crucial characteristics of the dominant modes [11], [12], [66]. In order to measure the data of F , the time stamp of an individual PMU is 100 ms [5]. The subspace-based statespace system identification (4SID) proposed in the previous work [11] has been applied to analyze characteristics of the FO, i.e., damping and frequency. In this paper, the 4SID is implemented by Signal Processing Toolbox [69] and System Identification Toolbox [70].
To analyze effects of the FO in a power system with CCGs, four different case studies are given as follows: Case Study 1: The FDs with 0.215, 0.385, and 0.442 Hz sine components are injected into the converter control loop of either WG-41, WG-31, or SPV-52. Thus, these CCGs behave as the FO sources. The penetration level of CCGs (defined as PLV in Fig. 5) is adjusted by two levels, i.e., 24% and 36%. Fig. 5 shows the maximum tie-line amplitude (P max tie ) resulted by such scenarios. As can be observed, P max tie tends to increase when the FO frequency (F d ) is close to 0.215, 0.385, and 0.442 Hz (dominant modes). Besides, the scenario is intensified by the increased penetration level of CCGs (or high PLV ). Case Study 2: The FDs with 0.215, 0.385, and 0.442 Hz sine components are injected into different CCGs, while the FD amplitude (P d ) varies from small to large values. From the results in Fig. 6, it is evident that the FOs caused by different types of CCGs can lead to high oscillation amplitude, especially when P d is increased. At P d = 0.3 pu, the system severely oscillates and eventually becomes unstable.
Case Study 3: The same FDs in Case Study 2 with fixed 0.1 pu P d are applied at WG-52 (Area-5). Besides, the total inertia of Area-5 is assumed to decrease from 100% to 60% with 20% step. Fig. 7 demonstrates the system responses for this case. Although P d is small as 0.1 pu, the system can become unstable when the total inertia is reduced. Moreover, the conventional SG replacement by CCGs can result in the decreasing of total system inertia. This scenario could happen in future power systems with high penetration of CCGs.
Case Study 4: To verify the result in a wide range of system operating conditions and various uncertainties, the probabilistic analysis is applied to analyze the impact of FO by randomly and simultaneously changing possible system conditions, i.e., location and duration of the FD, size of the CCGs, penetration levels of CCGs, and total inertia of the system. As a result, by such random conditions, the probability of damping ratios under 1,000 different operating points is depicted in Fig. 8. It can be observed that the FO caused by RESs exhibits both critical and negative damping, especially  when the frequency of FD is in the range of inter-area mode. During the FO, damping of the third inter-area mode is below the industry standard (3%). Therefore, it is evident that the FO from RESs significantly degrades the damping of dominant mode. Consequently, it may result in small-signal instability.
When the FD contaminates in wind speed or solar radiation, the CCGs can become the possible FO sources; the severe FO might occur in well-damped power systems. This could degrade the power system stability for certain conditions such as large FD amplitude, adjacent of frequencies of FO and inter-area modes, and FO in systems with low inertia. When the frequency of FD is in the inter-area oscillation frequency band, the resonance occurs, and the damping ratio of the dominant mode is significantly decreased (approximately zero). Conversely, if the frequency of FD is nearly equal to that of the dominant mode, it may result in negative damping and consequently leads the system to the unstable region.

VI. CONCLUSION AND FUTURE RESEARCH DIRECTIONS
An overview of the FO has been given with representative case studies. A thorough review of recent methods of FO analysis, detection, source identification, and mitigation has been discussed in this paper. The problem of FO regarding the penetration of CCGs in future power systems has been introduced. Besides, simulation results demonstrate that the FO caused by CCGs can potentially lead to power system instability. The results also reveal that the system with a high FD amplitude, proximity to inter-area mode frequencies, and low system inertia may be susceptible to the FO. The gaps and future research directions are given as follows: 1) FO analysis, detection, and source identification in power systems with CCGs were not well studied or understood. Most of the literature attempted to analyze the FO, which is originated from SGs. In future power systems, uncertainties of CCGs may affect the accuracy and reliability of such means, and they may result in false alarms of the FO. Therefore, it is important to consider all possible uncertainties that might cause errors in these processes; 2) Using network solutions to mitigate the FO (i.e., disconnection of FO sources and/or reduction in the output from the FO source) in low-inertia power systems with CCGs may ignite other instability issues. To suppress the FO effectively, CCGs with FO damping controllers and/or coordinated control with PSS in SGs and/or FACTS devices will be an emerging field of future research; 3) As reported by [13], a control solution is an alternative option to mitigate the FO in future power systems effectively without locating the source of the original FO. However, control solutions may not entirely mitigate the sustained FO. Therefore, a coordination of control and network solutions will be an interesting research topic; 4) Control solutions may reduce the FO amplitude in the measured signal. Therefore, the observability of FO could be decreased. Consequently, the accurate source identification could be limited due to the obscured FO source. Hence, the interaction between control and network solutions should be taken into account; 5) With the penetration of CCGs in future power systems, this scenario may lead to greater complexity in locating and ranking the FO sources, difficulty to alleviate the FO. Consequently, the possibility of new control strategies of CCGs and/or SGs will be introduced to mitigate the FO. Furthermore, the following analyses, enhancements, and modifications are promising: RAKIBUZZAMAN SHAH (Member, IEEE) received the Ph.D. degree from The University of Queensland, Brisbane, QLD, Australia. He is currently a Senior Lecturer in smart power systems engineering with the School of Engineering, Information Technology and Physical Sciences, Federation University Australia (FedUni Australia). Prior to join Federation University Australia, he worked with The University of Manchester, The University of Queensland, and Central Queensland University. He has experience working at, and consulting with, DNOs and TSOs on individual projects and collaborative work on large projects (EPSRC project on multi-terminal HVDC, Scottish and Southern Energy Multi-infeed HVDC) -primarily on the dynamic impact of integrating new technologies and power electronics into large systems. He is an Active Member of the CIGRE. He has more than 80 international publications. He has also spoken at the leading power system conferences around the world. His research interests include future power grids (i.e., renewable energy integration, and wide-area control), asynchronous grid connection through VSC-HVDC, power system dynamics and stability, application of data mining in power systems, application of control theory in power systems, distribution system energy management, and low carbon energy systems. Prior to joining The University of Queensland, he was attached to energy field of study at AIT. His previous professional positions include Planning Engineer at Generation Planning Division for two years at Ceylon Electricity Board, Sri Lanka, and a Project Leader at the Centre of Excellence in Electric Power Technology for one year at Chulalongkorn University, Thailand. He has been the Director of Higher Degree Research Training and Post Graduate Coordinator at the School of Information Technology and Electrical Engineering, The University of Queensland, since July 2019. His research interests include analytical studies on electric power grids, power system stability and dynamics, grid integration of renewable energy, battery energy storage, and electric vehicle integration to power systems.