Study on a Correlation-Based Anti-Islanding Method under Wider Frequency Trip Settings for Distributed Generation

Islanding phenomenon of distributed generation (DG), such as photovoltaic (PV) generation, is undesirable because it causes safety issues for utility service personnel and power system equipment. Many anti-islanding methods have been studied since DG appeared in electric power systems (EPSs). Most anti-islanding methods focus on disconnecting DG from the grid using functionality to detect islanding under narrow frequency trip settings, because safety issues have a higher priority. However, as DG plays a key part of an EPS, a significant loss of DG due to a short disturbance could result in a reliability issue for the EPS. Corresponding to this matter, new international standards, such as IEEE standard 1547–2018, require more sophisticated and complex functionalities for grid-connected DGs by adopting ride-through technologies and wider voltage/frequency trip settings. Since most anti-islanding functions of inverter-based DG have been based on the frequency of the inverter voltage, it is more difficult to detect islanding under wider frequency trip settings. This paper presents a correlation-based anti-islanding method (AIM) without depending on the frequency trip of inverter-based DGs. Simulation results are provided to verify the performance of the correlation-based anti-islanding method. As a result, the proposed method detects islanding at 0.116 s under wider frequency trip setting by the IEEE Std. 1547–2018 test condition, while the popular active frequency drift method with positive feedback does not detect islanding using the same current disturbance.


Introduction
Distribution generation (DG) based on renewable energy resources provides power to local loads in coordination with the main power grid [1][2][3]. DG is now commonly installed in electric power systems (EPSs) [4]. At present, EPSs have become increasingly complex in terms of their operation and controls because of the significant penetration of DGs [5]. The islanding phenomenon of DG has been regarded as one of the major issues of increased DGs in EPSs. Islanding is defined as a condition in which a portion of the utility with a DG and load remains energized even though that portion is disconnected from the remainder of the utility system, as shown in Figure 1 [6]. Since islanding causes safety issues for both utility service personnel and power system equipment, it is desirable that it is prevented [7]. However, as the utilization of DG has increased rapidly in EPSs, a possible significant loss of DG due to a short disturbance could potentially aggravate the problem and cause a wide-spread blackout.
balance between DG and local loads in the form of frequency drifting [15,16] and reactive power drifting [17,18]. In addition, neural network, pattern recognition and machine learning-based AIMs have been proposed [19,20]. Since most anti-islanding functions of inverter-based DG have been based on the frequency of the inverter voltage, it is more difficult to detect islanding under wider frequency trip settings. To solve this technical barrier, this paper is based on a correlation technique [21].  In this paper, study of a correlation-based AIM under wider frequency trip settings for distributed generation is presented. A non-detection zone (NDZ) by frequency trip settings is analyzed under IEEE Std. 1547-2018, compared with the previous settings under IEEE Std. 1547Std. -2003. Then, a newly proposed correlation-based AIM method is presented with a corresponding simulation for its verification.

NDZ Comparisons
In the early days of the dissemination of DG, safe energy generation of the EPS was important. For voltage and frequency variation, which are indicators of EPS accidents such as blackouts, the focus was on disconnecting the DG from the EPS. As DG has been implemented in EPSs with large As a result of this issue, new requirements for grid-connected technology, including ride-through and anti-islanding technology, which are more sophisticated and complex, have been recently proposed [8,9]. According to these new requirements, ride-through is the ability to withstand voltage and frequency disturbances, whereas trip is the cessation of output without immediate return to service. As trip requirements are related to the islanding of DG, a requirement for anti-islanding within two seconds is applied when DG is isolated from the upstream part of the area EPS. To date, as shown in Figure 2, a variety of anti-islanding methods (AIMs), including local techniques and remote techniques, have been developed and implemented. Remote methods depend on communication techniques between DG and the EPS but tend to be more costly than local methods [10]. Local methods are usually implemented into the grid-connected inverter of DG and they are typically classified as passive and active methods. Passive methods, based on power mismatches between DG power and local loads, detect islanding only by monitoring several system parameters such as voltage magnitude [11], frequency [12], phase jump [13], and harmonic detection [14]. On the other hand, active islanding methods perturb the DG inverter current in order to break the power balance between DG and local loads in the form of frequency drifting [15,16] and reactive power drifting [17,18]. In addition, neural network, pattern recognition and machine learning-based AIMs have been proposed [19,20]. Since most anti-islanding functions of inverter-based DG have been based on the frequency of the inverter voltage, it is more difficult to detect islanding under wider frequency trip settings. To solve this technical barrier, this paper is based on a correlation technique [21].
Appl. Sci. 2020, 10, x 2 of 11 As a result of this issue, new requirements for grid-connected technology, including ridethrough and anti-islanding technology, which are more sophisticated and complex, have been recently proposed [8,9]. According to these new requirements, ride-through is the ability to withstand voltage and frequency disturbances, whereas trip is the cessation of output without immediate return to service. As trip requirements are related to the islanding of DG, a requirement for anti-islanding within two seconds is applied when DG is isolated from the upstream part of the area EPS. To date, as shown in Figure 2, a variety of anti-islanding methods (AIMs), including local techniques and remote techniques, have been developed and implemented. Remote methods depend on communication techniques between DG and the EPS but tend to be more costly than local methods [10]. Local methods are usually implemented into the grid-connected inverter of DG and they are typically classified as passive and active methods. Passive methods, based on power mismatches between DG power and local loads, detect islanding only by monitoring several system parameters such as voltage magnitude [11], frequency [12], phase jump [13], and harmonic detection [14]. On the other hand, active islanding methods perturb the DG inverter current in order to break the power balance between DG and local loads in the form of frequency drifting [15,16] and reactive power drifting [17,18]. In addition, neural network, pattern recognition and machine learning-based AIMs have been proposed [19,20]. Since most anti-islanding functions of inverter-based DG have been based on the frequency of the inverter voltage, it is more difficult to detect islanding under wider frequency trip settings. To solve this technical barrier, this paper is based on a correlation technique [21].  In this paper, study of a correlation-based AIM under wider frequency trip settings for distributed generation is presented. A non-detection zone (NDZ) by frequency trip settings is analyzed under IEEE Std. 1547-2018, compared with the previous settings under IEEE Std. 1547-2003. Then, a newly proposed correlation-based AIM method is presented with a corresponding simulation for its verification.

NDZ Comparisons
In the early days of the dissemination of DG, safe energy generation of the EPS was important. For voltage and frequency variation, which are indicators of EPS accidents such as blackouts, the focus was on disconnecting the DG from the EPS. As DG has been implemented in EPSs with large In this paper, study of a correlation-based AIM under wider frequency trip settings for distributed generation is presented. A non-detection zone (NDZ) by frequency trip settings is analyzed under IEEE Std. 1547-2018, compared with the previous settings under IEEE Std. 1547-2003. Then, a newly proposed correlation-based AIM method is presented with a corresponding simulation for its verification.

NDZ Comparisons
In the early days of the dissemination of DG, safe energy generation of the EPS was important. For voltage and frequency variation, which are indicators of EPS accidents such as blackouts, the focus was on disconnecting the DG from the EPS. As DG has been implemented in EPSs with large capacities, a significant loss of DG due to a short disturbance could cause another reliability issue for the EPS. As a result, new international standards, such as IEEE standard 1547-2018, require more sophisticated and complex requirements for grid-connected DGs. For short voltage and frequency disturbances, the ride-through capability was adopted to continue the operation of DG, as shown in Figure 3. These wider voltage and frequency trip windows allow DG to generate electric power even though there is a slight voltage and frequency disturbance. On the other hand, islanding of DG should be prevented in an EPS because it may cause safety issues. Thus, IEEE Std. 1547-2018 requires islanding detection within 2 s without compromise. However, since most anti-islanding methods are based on frequency trip windows, there have been concerns about the reduced effectiveness of anti-islanding under the new standard requirements.
Appl. Sci. 2020, 10, x 3 of 11 capacities, a significant loss of DG due to a short disturbance could cause another reliability issue for the EPS. As a result, new international standards, such as IEEE standard 1547-2018, require more sophisticated and complex requirements for grid-connected DGs. For short voltage and frequency disturbances, the ride-through capability was adopted to continue the operation of DG, as shown in Figure 3. These wider voltage and frequency trip windows allow DG to generate electric power even though there is a slight voltage and frequency disturbance. On the other hand, islanding of DG should be prevented in an EPS because it may cause safety issues. Thus, IEEE Std. 1547-2018 requires islanding detection within 2 s without compromise. However, since most anti-islanding methods are based on frequency trip windows, there have been concerns about the reduced effectiveness of anti-islanding under the new standard requirements.
The relationship between reactive power mismatch ΔQ and frequency threshold can be obtained by [22]: where Pinv is the active power of the DG inverter; ΔQ is the reactive power flow from the grid; f is the grid frequency; fmin and fmax are the under and over frequency thresholds, respectively; and Qf is a quality factor. The quality factor is defined as the measure of the resonance strength of the islanding test load, as shown in Figure 1a [23]. The relationship between reactive power mismatch ∆Q and frequency threshold can be obtained by [22]: where P inv is the active power of the DG inverter; ∆Q is the reactive power flow from the grid; f is the grid frequency; f min and f max are the under and over frequency thresholds, respectively; and Q f is a quality factor. The quality factor is defined as the measure of the resonance strength of the islanding test load, as shown in Figure 1a [23].
With the limit values of the frequency specified in Table 1, the corresponding NDZ was derived. This quantitative NDZ is the basis to calculate the amount of frequency perturbation of the DG inverter, which causes the harmonic distortion of the current. It has been one of the main research targets for detecting islanding with high power quality and low harmonics.
As shown in Table 1, the islanding test conditions changed to reduce the effect on the EPS by decreasing Q f from 2.5 to 1. Additionally, the frequency relay settings were changed in a wider direction to prevent nuisance trips. Table 1 shows that NDZ varies by Equation (1) according to the set value of the frequency trip settings and the value of Q f . Among the NDZ ranges of the three different standards, IEEE Std. 1547-2018 shows the largest NDZ because it has the widest frequency trip settings. Under the wide range of frequency trip settings, the current disturbance command of AIM must increase to make the frequency move out of the range when islanding occurs. Since this disturbance appears as the current distortion, harmonic components increase or displacement power factor decreases, resulting in poor power quality. Thus, this paper presents a correlation-based AIM that does not depend on the frequency trip.

The Proposed AIM
As one of the most used AIMs, the active frequency drift (AFD) method with positive feedback has been studied often [24][25][26]. Accordingly, this method was applicable to be implemented in a commercial photovoltaic (PV) inverter to meet the requirements of IEEE Std. 1547-2003. As shown in Figure 4, the AFD method uses the chopping fraction, cf, as a frequency perturbation parameter, which is defined as the ratio of zero time (t z ) to half of the period of voltage waveform (T 1 /2) as follows: where t z is zero time of the inverter current and T 1 is a period of voltage waveform.
direction to prevent nuisance trips. Table 1 shows that NDZ varies by Equation (1) according to the set value of the frequency trip settings and the value of Qf. Among the NDZ ranges of the three different standards, IEEE Std. 1547-2018 shows the largest NDZ because it has the widest frequency trip settings. Under the wide range of frequency trip settings, the current disturbance command of AIM must increase to make the frequency move out of the range when islanding occurs. Since this disturbance appears as the current distortion, harmonic components increase or displacement power factor decreases, resulting in poor power quality. Thus, this paper presents a correlation-based AIM that does not depend on the frequency trip.

The Proposed AIM
As one of the most used AIMs, the active frequency drift (AFD) method with positive feedback has been studied often [24][25][26]. Accordingly, this method was applicable to be implemented in a commercial photovoltaic (PV) inverter to meet the requirements of IEEE Std. 1547-2003. As shown in Figure 4, the AFD method uses the chopping fraction, cf, as a frequency perturbation parameter, which is defined as the ratio of zero time (tz) to half of the period of voltage waveform (T1/2) as follows: where tz is zero time of the inverter current and T1 is a period of voltage waveform. Without the AFD method, the current is controlled to be in phase with the voltage waveform, which results in a unity power factor control. When the AFD method is applied, the frequency of the DG inverter is maintained by the grid voltage when the grid is connected, so there is no frequency variation due to zero time of the AFD reference current. It is important to design an appropriate cf value because harmonic components are generated by the injected AFD current command. When the DG inverter is disconnected from the grid, the AFD current command makes the frequency drift to the frequency trip set value, and the inverter stops [24][25][26].
If an AFD current disturbance with a fixed cf can be canceled by local load adjustment, the chopping fraction is usually modified with a positive feedback as follows [24]: where k is the present status, Kp is a proportional coefficient, and Freq is the measured frequency of the voltage.  Without the AFD method, the current is controlled to be in phase with the voltage waveform, which results in a unity power factor control. When the AFD method is applied, the frequency of the DG inverter is maintained by the grid voltage when the grid is connected, so there is no frequency variation due to zero time of the AFD reference current. It is important to design an appropriate cf value because harmonic components are generated by the injected AFD current command. When the DG inverter is disconnected from the grid, the AFD current command makes the frequency drift to the frequency trip set value, and the inverter stops [24][25][26].
If an AFD current disturbance with a fixed cf can be canceled by local load adjustment, the chopping fraction is usually modified with a positive feedback as follows [24]: where k is the present status, K p is a proportional coefficient, and Freq is the measured frequency of the voltage. Since the displacement factor (DPF) is 1, the power factor (PF) and the reactive power can be expressed as follows [27]: Since total harmonic distortion of current (THD i ) is directly proportional to cf, the amount of reactive power variation (Q/P) in Equation (6) can be determined by cf. The previous research showed the relationship between cf and THD i through some experiments [27]. The allowable limit of THDi (5%) by the standard sets the maximum cf about 5% [27,28]. In the case of the new IEEE Std. 1547-2018, the amount of perturbation (∆Q/P inv ) increases dramatically, as shown in Table 1. Consequently, a higher cf of the AFD method is required to prevent islanding. Thus, it is difficult to use the conventional AFD with positive feedback (AFDPF) by IEEE Std. 1547-2018 because the corresponding THD i is around 12.77% which is higher than the standard limit [28]. Since NDZs, as shown in Table 1, are valid only when AIMs depend on the frequency trip settings, this paper presents a novel method using a correlation parameter without depending on the frequency trip settings.
where k is the present status and N is the number of line frequency for C p calculation. As shown in Equation (7), this paper presents the correlation parameter C p as an islanding detection indicator without depending on the frequency trip setting. The proposed method uses the fact that the frequency of the DG inverter voltage has a strong correlation with the current frequency deviation after islanding. As shown in Figure 5, when the grid is connected, the inverter voltage of the DG is dominated by the rigid grid voltage source and has a weak correlation with the proposed AFD frequency perturbation. Thus, C P is maintained to be almost zero before t 0 . On the other hand, after islanding, C P starts to increase after t 0 and reaches C P,set , and islanding is prevented.
Appl. Sci. 2020, 10, x 5 of 11 Since the displacement factor (DPF) is 1, the power factor (PF) and the reactive power can be expressed as follows [27]: Since total harmonic distortion of current (THDi) is directly proportional to cf, the amount of reactive power variation (Q/P) in equation (6) can be determined by cf. The previous research showed the relationship between cf and THDi through some experiments [27]. The allowable limit of THDi (5%) by the standard sets the maximum cf about 5% [27,28]. In the case of the new IEEE Std. 1547-2018, the amount of perturbation (ΔQ/Pinv) increases dramatically, as shown in Table 1. Consequently, a higher cf of the AFD method is required to prevent islanding. Thus, it is difficult to use the conventional AFD with positive feedback (AFDPF) by IEEE Std. 1547-2018 because the corresponding THDi is around 12.77% which is higher than the standard limit [28]. Since NDZs, as shown in Table 1, are valid only when AIMs depend on the frequency trip settings, this paper presents a novel method using a correlation parameter without depending on the frequency trip settings.
where k is the present status and N is the number of line frequency for Cp calculation. As shown in Equation (7), this paper presents the correlation parameter Cp as an islanding detection indicator without depending on the frequency trip setting. The proposed method uses the fact that the frequency of the DG inverter voltage has a strong correlation with the current frequency deviation after islanding. As shown in Figure 5, when the grid is connected, the inverter voltage of the DG is dominated by the rigid grid voltage source and has a weak correlation with the proposed AFD frequency perturbation. Thus, CP is maintained to be almost zero before t0. On the other hand, after islanding, CP starts to increase after t0 and reaches CP,set, and islanding is prevented.  For the AFDPF method, the key design parameter of cf is the positive feedback gain K p , which accelerates cf deviation as frequency changes. K p was designed and chosen by trial-and-error simulations for short frequency trip settings, 59.3 and 60.5 Hz from IEEE Std. 1547-2003. For IEEE Std. 1547-2018, the frequency trip settings, 56.5 and 62 Hz, are wider. Thus, if AFDPF method is used, K p for IEEE Std. 1547-2018 should be larger than K p for IEEE Std. 1547-2003, which increases the harmonic components. In this paper, K p 0.02, which is designed for 59.3 and 60.5 Hz, is also used to show the feasibility for 56.5 and 62 Hz by using the proposed AIM. In summary, the proposed method does not depend on the frequency trip settings and is based on the correlation between the rate of change of the input current perturbation and the rate of change of the corresponding frequency. This means that islanding can be detected using the proposed correlation parameter with less cf perturbation under IEEE Std. 1547-2018.

Simulation Results
The performance of the proposed method was verified using PSIM software under the conditions of IEEE Std. 1547-2018. The key parameters for islanding test are shown in Table 2 and simulation circuit is shown in Figure 6. The local load was chosen as a unity power quality factor for a 3 kW DG inverter system. By adjusting tunable RLC local load, a test condition was implemented for the worst case, the power balance condition between DG generation and local load consumption. The threshold value of the correlation factor was chosen to be 0.0013, which was derived from the value of cf for IEEE Std. 1547-2003. This is to show that the designed cf for IEEE Std. 1547-2003 can also be applied to IEEE Std. 1547-2018 using the proposed method. In other words, the lower cf value, which was designed for IEEE Std. 1547-2003, can be used for IEEE Std. 1547-2018.  Figure 7 shows islanding test results when no AIM is applied. When the local loads are almost matched with DG power generation and the grid is disconnected at 0.5 s, the measured frequency is almost maintained to be 60 Hz. Thus, islanding is not detected at all. Figure 8 shows anti-islanding test results when the conventional AFD method with 5% cf is used. Before islanding occurs, the inverter output current is clearly distorted by cf, as shown in Figure 8. Once islanding occurs, the frequency of inverter voltage stays still around 60 Hz, because DG source and load power are balanced under the constant cf disturbance. Thus, islanding is not prevented in this case. In the same way, Figure 9 shows anti-islanding test results when AFDPF method is used. Before islanding occurs, it is hard to recognize the distortion of the inverter output current as shown in Figure 9, because the initial value cf for AFDPF method is 1%. After islanding at 0.5 s, the frequency is drifted by cf over 60.5 Hz and then stabilized under 62 Hz. Even though cf is increased by the positive feedback, the frequency variation after islanding is not enough to detect islanding by depending on frequency trip settings. This means that AFDPF method may meet the requirements of anti-islanding under IEEE Std. 1547-2003, whose frequency trip setting is at 60.5 Hz. However, this AFDPF method does not meet the requirement of IEEE Std. 1547-2018 because its frequency trip point is 62 Hz. Figure 10 shows anti-islanding test results when the proposed method is used. Before islanding occurs, the distortion of the current is small like the AFDPF method since the initial value cf is 1%. In addition, before islanding, the proposed correlation parameter C P is maintained to be almost zero because frequency is dominated by the rigid grid voltage source, not the chopping fraction. In other words, there is little correlation between cf and frequency before islanding. However, after islanding, the frequency is determined by the chopping fraction with positive feedback. Therefore, the proposed correlation factor starts to increase to be over 0.0013. Therefore, islanding can be prevented by the proposed islanding indicator, C P .
Appl. Sci. 2020, 10, x 7 of 11          The simulation results are summarized in Table 3. When the grid is connected to DG, the THD i values of AIM methods are assessed. In the conventional AFD method, THD i increases to 5.53% as the cf value increases from 1% to 5%. Both the AFDPF method and the proposed method show low THD i , which is 3.18%, because the initial value cf 1% is only inserted when the grid is connected. For AFD method, islanding is not prevented because DG source and load power were balanced. For AFDPF method, islanding is also not prevented, because the frequency deviation is not enough by using low K p gain 0.02. In case of the proposed method, islanding is prevented with the same K p gain of AFDPF method, because this method does not depend on the frequency trip settings.

Conclusions
This paper presents a correlation-based AIM under wider frequency trip settings for DG. Since a significant loss of DG due to a short disturbance could cause another reliability issue for an EPS, wider frequency trip settings were introduced in IEEE Std. 1547-2018. Under these conditions, the conventional AIM might not detect islanding with lower current distortion. Thus, the proposed method presents an islanding detection indicator as the correlation parameter between current disturbance and the corresponding frequency, and shows high islanding-detection capability with less current distortion through simulation results. According to the simulation results, the proposed AIM method detects islanding under the IEEE Std. 1547-2018 test condition within 2 s, which meets the standard requirement, and it has around 3.18% THDi, which is relatively lower compared with the conventional AFD method having 5% cf.

Conclusions
This paper presents a correlation-based AIM under wider frequency trip settings for DG. Since a significant loss of DG due to a short disturbance could cause another reliability issue for an EPS, wider frequency trip settings were introduced in IEEE Std. 1547-2018. Under these conditions, the conventional AIM might not detect islanding with lower current distortion. Thus, the proposed method presents an islanding detection indicator as the correlation parameter between current disturbance and the corresponding frequency, and shows high islanding-detection capability with less current distortion through simulation results. According to the simulation results, the proposed AIM method detects islanding under the IEEE Std. 1547-2018 test condition within 2 s, which meets the standard requirement, and it has around 3.18% THDi, which is relatively lower compared with the conventional AFD method having 5% cf.