A new islanding detection technique based on rate of change of reactive power and radial basis function neural network for distributed generation

Abstract

Accurate and fast islanding detection of distributed generation is extremely important for its effective operation in distribution systems. For this purpose, several islanding detection methods have been suggested. Among them, hybrid islanding detection techniques are preferred due to their minimum effects on the power system. However, hybrid islanding detection techniques also suffer from two main limitations. They still degrade the power quality and also take comparatively large time to detect the islanding phenomenon. Thus, fast detection and power quality degradation issues are still not solved by hybrid islanding detection techniques. To address this issue, this paper suggests a new islanding detection technique based on rate of change of reactive power (ROCORP) and radial basis function neural network (RBFNN). The proposed technique uses ROCORP as the RBFNN input. The appropriate database of several islanding and non-islanding events is generated by performing the offline simulations on 26 Bus Malaysian distribution system for training the RBFNN. The simulation results shows that it can detects islanding and non-islanding events very fast without degrading the power quality of the system and is independent of threshold limitations.

Keywords

Islanding detection Distributed generation Mini-hydro ROCORP RBFNN

1 Introduction

In the recent years, the usage of Distributed Generation (DG) is inevitable due to exponential increased demand of electrical energy in the power system. High integration of DG resources in the power system has caused several challenging issues for their successful operation. The main technical issue in DG integration with main grid is Islanding condition [1]. To ensure the reliable operation of power system, IEEE standard 1547 has put the condition to recognize the islanding event and isolate the DG from distribution system within 2 seconds (100cycles) [2]. This requires an accurate and fast islanding detection technique for safe operation of DGs in distribution network [3].

Up to now, various methods have been suggested for the detection of an islanding event namely passive, active and hybrid islanding detection methods. In order to justify the need for a new technique, it is better to compare all above techniques in terms of certain parameters. The ability of an islanding detection technique is determined by following four parameters [4].

Non-detection Zone (NDZ): The non-detection zone is an important factor for the accuracy of an islanding detection method. It is defined as the range, (in terms of inequality between DG and load) where an islanding detection methods would not identify the islanding. Ideally, islanding detection techniques should have zero non detection zone. However, in practice, the islanding detection method with very small non-detection zone possesses least error as compared with huge non-detection zone (NDZ).

Threshold Setting: Most passive islanding detection methods uses network parameters such as active power, reactive power, frequency, voltage and compare these measured values with a certain threshold to discriminate in islanding and non-islanding cases. However, these methods require much attention in setting the threshold values since small threshold value may cause unusual shutdown but, huge threshold value may result in failure in islanding detection.

Power Quality Degradation: Most active islanding detection techniques injects high frequency signals at the fixed instance of time in the network for the detection of island. However, most of the times, these injection of signals is not necessary and seriously degrade the power quality.

Fast detection of Islanding Events: The detection of islanding phenomenon in short time is a primary step for secured operation and integration of DG resources downstream of protective devices.

Based upon above four parameters, passive techniques (rate of change of power (POCOP) [5], rate of change of frequency over power (ROCOFOP) [6], change of impedance [7], voltage unbalance [8], over/under (O/U) voltage and over/under frequency [8], phase jump detection [9], and harmonic distortion (voltage and current) [10] have the benefit that they do not degrade the power quality of the system. Still, these methods have the limitation that they encounter large non-detection zones and their performance is sensitive to threshold setting [11].

On the contrary, the active techniques such as fuzzy adaptive phase drift control [12], impedance measurement [13], Active frequency drift (AFD) [14], Slip-mode frequency shift (SMS) [15], frequency jump (FJ) [16], high frequency signal injection [17] and phase PLL perturbation method [18] have the advantage that in comparison to passive techniques, they have smallest non-detection zone [19]. However, these methods have resulted in two other issues of power quality degradation and taking large time to identify the islanding incident [20].

In order to address the issue of power quality degradation, hybrid islanding detection methods have been suggested. These methods combine the advantages of active and passive methods. In these methods, active methods are utilized if and only if when passive methods are unable to identify islanding incidents [21]. The algorithms of such methods includes rate of change of reactive power (ROCORP) and load connecting strategy [22], voltage unbalance/frequency shift and average rate-of-change of voltage/real power shift [23], voltage unbalance and high frequency (HF) impedance [24].

These technique have the benefit of getting very small NDZ, and has greatly reduced the power quality degradation because signals are not inserted to the network at each instance of time. However, due to combination of passive and active techniques, computational time to identify the islanding phenomenon is highly increased. Figure 1 shows the time taken by various hybrid techniques to detect the islanding phenomenon.

Fig.1

Islanding detection time of various hybrid techniques.

It can be observed from the Fig. 1 that the time taken by hybrid islanding detection techniques varies from 0.15 seconds to 0.8 seconds [22 , 25–33]. Moreover, the hybrid islanding detection techniques still do not solve the power quality degradation problem for the distribution network to some extent and suffer from threshold setting problem too. Thus, it may be noted that fast detection, power quality degradation, and threshold setting issues in islanding detection techniques are still not addressed by hybrid techniques. Thus, an effective islanding detection method is still essential, which may possess a smaller non detection zone (NDZ),fast detection of islanding phenomenon, free from power quality degradation, and independent of threshold setting limitation.

To address this issue, the research trend shift towards the computational intelligence based islanding detection methods. Many researchers have proposed islanding detection techniques using Adaptive Neuro-Fuzzy Inference system (ANFIS), Artificial Neural Network (ANN), Support Vector Machine (SVM), Probabilistic neural network (PNN), Decision Tree (DT), and Fuzzy Logic Control (FLC). However, it has been observed that various researchers have used several parameters with computational intelligence based techniques as shown in Fig. 2.

Fig.2

Number of parameters used in computational islanding detection methods.

It can be perceived from Fig. 2 that number of parameters varies from 3 to 12 in some papers that has enhanced the complexity of these techniques and outweighs their advantages [4 , 34–39].

Apart from this, many researchers have applied computational intelligence based techniques on active and hybrid islanding detection techniques as proposed in references [36 , 40–43]. This has resulted in too much complications and still encountered the limitations of threshold setting, and power quality degradation. However, it is believed that the execution of computational intelligence based methods with passive methods for islanding detection may have the ability to overcome all these drawbacks of non-detection zone, threshold setting problem, power quality degradation and large time detection. Based upon this idea, this paper suggests a computational intelligence based islanding detection method with passive parameter to overcome all these limitations.

2 Methodology

The Suggested islanding detection method is based on two main parts namely passive parameter and computational technique. Among various passive parameters, this paper considers rate of change of reactive power as the input variable for RBFNN. It is because, research has shown that among passive parameters, ROCORP is more sensitive compared to other parameters [22, 44]. Furthermore, Zhihong et al. [44] has proved that ROCORP is more sensitive than other passive parameters as shown in following equations $- 17.36 % \leq Δ P / P \leq 29.13 %$ (1) $- 5.94 % \leq Δ Q / Q \leq 4.11 %$ (2)

It can be noticed from Equation (1) that reactive power mismatch is most sensed (6%) as compared to active power mismatch (29%) [44]. The training data for RBFNN is obtained by measuring ROCORP for several islanding and non-islanding incidents by performing offline simulations on Malaysian Distribution Network in PSCAD/EMTDC software version 4.2.1. This paper uses RBFNN for classification purposed between islanding and non-islanding cases. RBFNN is a powerful tool, which is modeled for complex and nonlinear systems with least number of input and output incident for training purpose. The modeling details of test system, RBFNN and its training is described in next sections.

2.1 Modeling of Test System

The distribution network used in this paper is are al 11 kV Malaysian distribution network. The distribution network consists of two mini hydro type DG resources having capacity of 2 MVA each operating at voltage of 3.3 kV. The test model given in Fig. 3 is simulated in PSCAD/EMTDC software. The utility grid is coupled with distribution system through step-down transformer.

Fig.3

Test System.

Both mini hydro type DGs are connected to a step up transformers (3.3kV/11kV) rated 2 MVA each. Both mini hydro DG unit are modeled in PSCAD/EMTDC library by using standard model for Exciter (IEEE type AC1A), governor (electro hydraulic PID controller) and hydraulic turbine (non-elastic water column without surge tank).

2.2 Radial Basis Function Neural Network (RBFNN) As Classifier

RBFNN has a simple structure and good generalizing ability to make it useful tool for complex and nonlinear networks and comparing with feed forward networks (FFNs). Furthermore, mostly used back propagation (BP) training algorithm for FFNs is slowest process of training the network, mainly for large size data. Since RBFNN has the ability to linear optimization there parameters with the help of hidden neurons directly from the input data train the network, it is generally much faster than FFNs, to complete the training. RBFNN is pervade by a set of inputs and output layer, between the inputs and outputs layer there is a layer of processing elements called hidden layers. Finally, due to their simple topological arrangement, RBFNNs have the capability to reveal how learning proceeds in a precise fashion. The assembly of the RBFNN is automatically generated on the base of input data. Typical structure of RBF network is shown in Fig. 4.

Fig.4

Typical structure of RBFNN.

The activation function at hidden layer use Gaussian function and between hidden layer to output layer uses linear function. The output of the RBFNN can be expressed by: $F (x) = \sum_{j = 1}^{m} \underset{j}{w} \underset{j}{h} (x)$ (3) where, x(t) is the input element is fabricated to network at time t, F(x) is the predictable output from the network. w_j is the connecting weight between the hidden RBF element j and the output neuron. The activation function of the RBF element is generally the following Gaussian function: $h_{j} (x) = e^{- (x - c_{j})^{2} / 2 σ_{j}^{2}}$ (4) where, c_j is the center of the j^th RBF element, and σ_j represents the radius or width of this element.

2.3 Generation of Data for RBFNN Training

The data is generated by performing off-line simulations of different islanding and non-islanding incidents in the Test system. The distribution system is coupled with transmission network through Grid via main bus and islanding would be created by opening Grid circuit breaker. For each islanding and non-islanding event, only one parameter ROCORP (dQ/dt) is measured for islanding detection. Figure 5 presents the basic architecture of proposed islanding detection technique and its parameters specification are shown in Table 1.

Fig.5

Proposed Architecture of islanding detection technique.

Table 1

Specification of the RBFNN structure

Parameters	Value
No. of neuron in input layer	1
No. of neuron in hidden layer	50
No. of neuron in output layer	1
Neural Network model	Radial Basis Function
Maximum number of epochs	1000
Error Threshold	1e-20
Hidden layer transfer function	Radbas
Output layer transfer function	Purelin
Spread of RBF	0.01

2.4 RBFNN Training

The cases adopted to collect data are simulated for a number of islanding and non-islanding incidents. Non-islanding cases include Single line to ground (S.L.G) fault, Double Line to ground (D.L.G) fault, three phase (LLLG) fault, line to line (L.L) fault, load increment &decrement, DG tripping, Induction motor starting and capacitor switching cases at various buses of the distribution system at different loads ratings. For each case, value of dq/dt are measured at particular instant of time and data of 480 cases have been chosen consisting of various islanding and non-islanding cases. Figure 6 shows number of cases for each islanding and non-islanding incident.

Fig.6

Number of Cases for Islanding and Non-islanding Events.

Out of 480 cases, 370 cases are considered for RBFNN training and about 110 cases are used for RBFNN testing. The data is divided into Islanding & non-islanding cases by assigning single output value for each case; Unity for Islanding event and Zero for Non-Islanding event respectively.

As a whole, data for RBFNN is divided into two columns i.e. first column for input and second column for output. The output column contains 0 and 1 for non-islanding & islanding case respectively. In the end, data obtained from these all cases are trained and the performance of RBFNN based islanding detection technique is verified through simulation results for different cases in MATLAB.Due to learning capability of RBFNN, it considers second column as its outputs and first column as its inputs. The proposed RBFNN model is trained with least possible error method. Figure 7 shows the training results of the suggested RBFNN with least error.

Fig.7

Training error of proposed RBFNN based method.

3 Simulation results

3.1 Islanding at Small and large power mismatch

In this Section, two islanding events are simulated for large power mismatch of 1.0 MW& 1.0 Mvar and Small power mismatch of 0.001 MW& 0.001 Mvar between generation and load at t = 0.3 s by opening the grid circuit breaker. The value of ROCORP for these cases are shown in Fig. 8.

Fig.8

ROCORP values and of proposed RBFNN for Large and small power mismatch.

At this instant ROCORP is measured and sent to the proposed RBFNN based islanding detection technique to differentiate between islanding and non-islanding incident. It can be observed from Fig. 8 that suggested RBFNN based islanding detection has accurately identified this event as islanding incident. It can be noticed from Fig. 8 that suggested method has successfully detected both of the cases of islanding at small & large power mismatch cases.

3.2 Non-Islanding cases

In this section, the performance of suggested RBFNN based islanding detection technique has been investigated for non-islanding incidents such as load variations (A), starting effect of induction motor (B), DG tripping (C), fault cases (D) and capacitor switching cases (E). To simulate load variation case, the load variation of 0.1 MW and 0.1 Mvar, is considered whereas for induction motor starting, 0.5 MW rating is considered. The tripping of DG may also occur because distribution system consists of two DG.

Furthermore, capacitor switching of 0.1 Mvar is also simulated to analyze the effectiveness of suggested method in distinguishing these events from islanding. All these incidents are simulated at the simulation time t = 0.3 s. The value of ROCORP for load variation case, induction motor starting case, DG tripping case, fault case and capacitor switching cases are shown in Fig. 9.

Fig.9

ROCORP values for various Non-islanding cases and response of the proposed RBFNN method.

The response of proposed RBFNN based technique for all these cases is shown in Fig. 9. It can be observed from Fig. 9 that proposed technique has successfully detected all these events as non-islanding event due to the fact that the output of the RBFNN is zero which is pre-defined condition of non-islanding.

4 Testing of Proposed RBFNN based Technique for 110 cases:

It shows the effectiveness of suggested method, its response for all 110 cases has been investigated. Among 110 cases that is 30% of the total data and consists of half of the islanding and non-islanding cases. Figure 10 shows the response of the suggested RBFNN based islanding detection technique for all 110 cases. Figure 10 shows that proposed RBFNN based technique has correctly differentiated between islanding and non-islanding incidents. However, it may be noticed that the response of few cases is not purely binary number (0 or 1) and lies in the range of –0.1 to 0.1 for non-islanding events and between 0.9 to 1.1 for islanding cases. Therefore, in order to improve the result to be exactly one or zero, “round function” of the MATLAB is used and the response of proposed technique for all these cases is again tested. Figure 11 shows the response of proposed technique with ”round function”.

Fig.10

Response of proposed RBFNN based technique for all 110 cases.

Fig.11

Response of Proposed Technique for all 110 cases with round function.

It can be observed from Fig. 11 that by using round function of the MATLAB, the proposed RBFNN based islanding detection technique has accurately distinguished between all these islanding and non-islanding incidents.

5 Discussions

This paper has presented the simulation results of the suggested technique for various islanding and non-islanding events to validate its performance. To highlight its effectiveness, the response of the proposed islanding detection technique is compared with other existing techniques in terms of time taken to detect the islanding event as shown in Table 2.

Table 2
Comparison of proposed islanding detection technique with hybrid islanding detection technique

Author Detection Time

Proposed method 0.02 s

Menon and Nehrir [33] method 0.17 s

Pigazo et al. [32] method 0.15 s

Rostami et al. [31] method 0.25 s

Shen et al. [30] method 0.2 s

Dhar and Dash [29] method 0.2 s

Garmrudi et al. [28] method 0.8 s

Mahat et al. [27] method 0.5 s

Kothari and Fernandes [26] method 0.412s

Khodaparastan et al. [25] method 0.5 s

Laghari et al. [22] method 0.44 s

Author	Detection Time
Proposed method	0.02 s
Menon and Nehrir [33] method	0.17 s
Pigazo et al. [32] method	0.15 s
Rostami et al. [31] method	0.25 s
Shen et al. [30] method	0.2 s
Dhar and Dash [29] method	0.2 s
Garmrudi et al. [28] method	0.8 s
Mahat et al. [27] method	0.5 s
Kothari and Fernandes [26] method	0.412s
Khodaparastan et al. [25] method	0.5 s
Laghari et al. [22] method	0.44 s

It can be noticed from the Table 2 that proposed islanding detection technique takes only 0.02 seconds to detect the islanding event as compared to other techniques. Furthermore, it can be noticed that proposed method has not degraded the power quality of the distribution system and not used any threshold setting. Thus, proposed RBFNN based islanding detection technique is a feasible option and has successfully addressed all issues of the existing techniques.

6 Conclusion

This paper has proposed a new fast and accurate islanding detection method based on ROCORP and RBFNN. The appropriate database of several islanding and non-islanding events has been obtained by performing the offline simulations on 26 Bus Malaysian distribution system for RBFNN training. The simulation results have shown that the proposed technique can accurately differentiate between islanding and non-islanding cases, including load variation, DG tripping, induction motor starting, faults, and capacitor switching cases. Furthermore, the proposed technique can identify islanding within one cycle, free from power quality degradation and threshold setting limitation. Hence, the proposed islanding detection technique has successfully solved the problem of non-detection zone, power quality degradation and threshold setting limitation, and fast detection which make it more appropriate for real time application.

References

Azim

, Zhu

, Saleem

H.A.

, Sun

, Li

, Shi

, et al., A decision tree based approach for microgrid islanding detection. Innovative Smart Grid Technologies Conference (ISGT), 2015 IEEE Power & Energy Society: IEEE; 2015. pp. 1–5.

Basso

, DeBlasio

, IEEE Smart Grid Series of Standards IEEE 2030 (Interoperability) and IEEE 1547 (Interconnection) Status: Preprint. National Renewable Energy Laboratory (NREL), Golden, CO.; 2012.

Jia

, Bi

, Liu

and Thomas

, Goodman A. Advanced islanding detection utilized in distribution systems with DFIG, International Journal of Electrical Power & Energy Systems 63 (2014), 113–123.

Al-Bataineh

and Kavasseri

R.G.

, Islanding Detection with Data Mining Methods-A Comparative Study. Green Technologies Conference (GreenTech), 2017 Ninth Annual IEEE: IEEE; 2017. pp. 104–109.

KNEK

Ahmad

, Selvaraj

and Rahim

N.A.

, A review of the islanding detection methods in grid-connected PV inverters, Renewable and Sustainable Energy Reviews 21 (2013), 756–766.

Pai

F-S.

and Huang

S-J.

, A detection algorithm for islanding-prevention of dispersed consumer-owned storage and generating units, IEEE Transactions on Energy Conversion 16 (2001), 346–351.

Kane

P.O.

and Fox

, Loss of mains detection for embedded generation by system impedance monitoring, Sixth International Conference on Developments in Power System Protection (Conf Publ No 434) 1997. pp. 95–98.

Jang

S-I.

and Kim

K-H.

, An islanding detection method for distributed generations using voltage unbalance and total harmonic distortion of current, IEEE Transactions on Power Delivery 19 (2004), 745–752.

Singam

and Hui

L.Y.

, Assessing SMS and PJD schemes of anti-islanding with varying quality factor. Power and Energy Conference, 2006 PECon’06 IEEE International: IEEE; 2006. pp. 196–201.

10.

Raza

, Mokhlis

, Arof

, Laghari

and Wang

, Application of signal processing techniques for islanding detection of distributed generation in distribution network: A review, Energy Conversion and Management 96 (2015), 613–624.

11.

Zeineldin

and Kirtley

J.L.

, A simple technique for islanding detection with negligible nondetection zone, IEEE Transactions on Power Delivery 24 (2009), 779–786.

12.

Shi

and Wu

, An islanding detection algorithm based on fuzzy adaptive phase drift control. Information and Automation (ICIA), 2013 IEEE International Conference on: IEEE; 2013. pp. 225–229.

13.

Mahat

, Chen

and Bak-Jensen

, Review on islanding operation of distribution system with distributed generation. Power and Energy Society General Meeting, 2011 IEEE: IEEE; 2011. pp. 1–8.

14.

Chen

, Wang

, Zhu

and Zhao

, An improved active frequency drift islanding detection method with lower total harmonic distortion, 2013 IEEE Energy Conversion Congress and Exposition 2013. pp. 5248–5252.

15.

Liu

, Kang

, Zhang

, Duan

and Lin

, Improved SMS islanding detection method for grid-connected converters, IET Renewable Power Generation 4 (2010), 36–42.

16.

, Cao

, Kuang

, Zeng

and Fang

, A review of islanding detection methods for microgrid, Renewable and Sustainable Energy Reviews 35 (2014), 211–220.

17.

Reigosa

, Briz

, Blanco

, García

and Guerrero

J.M.

, Active islanding detection for multiple parallel-connected inverter-based distributed generators using high-frequency signal injection, IEEE Transactions on Power Electronics 29 (2014), 1192–1199.

18.

Velasco

, Trujillo

, Garcera

and Figueres

, An active anti-islanding method based on phase-PLL perturbation, IEEE Transactions on Power Electronics 26 (2011), 1056–1066.

19.

Laghari

J.A.

, Mokhlis

, Karimi

, Bakar

A.H.A.

and Shahriari

, Artificial neural network based islanding detection technique for mini hydro type distributed generation, 3rd IET International Conference on Clean Energy and Technology (CEAT) 2014 2014. pp. 1–6.

20.

Khamis

, Shareef

, Bizkevelci

and Khatib

, A review of islanding detection techniques for renewable distributed generation systems, Renewable and Sustainable Energy Reviews 28 (2013), 483–493.

21.

Laghari

, Mokhlis

, Karimi

, Bakar

and Mohamad

, Computational Intelligence based techniques for islanding detection of distributed generation in distribution network: A review, Energy Conversion and Management 88 (2014), 139–152.

22.

Laghari

J.A.

, Mokhlis

, Bakar

A.H.A.

and Karimi

, A new islanding detection technique for multiple mini hydro based on rate of change of reactive power and load connecting strategy, Energy Conversion and Management 76 (2013), 215–224.

23.

Azim

, Li

and Zhao

, A hybrid islanding detection technique for inverter based distributed generations, Electrical Power and Energy Conference (EPEC), 2015 IEEE: IEEE; 2015. pp. 239–43.

24.

Huayang

, Zuyuan

, Long

and Yuan

, Based on hybrid algorithm of distributed power islanding detection, Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC), 2016 IEEE: IEEE; 2016. pp. 1360–1364.

25.

Khodaparastan

, Vahedi

, Khazaeli

and Oraee

, A novel hybrid islanding detection method for inverter-based DGs using SFS and ROCOF, IEEE Transactions on Power Delivery 32 (2017), 2162–2170.

26.

Kothari

and Fernandes

, A hybrid active islanding detection technique for single-phase inverter-based distributed generation system. Industrial Technology (ICIT), 2015 IEEE International Conference on: IEEE; 2015. pp. 2828–2834.

27.

Mahat

, Chen

and Bak-Jensen

, A hybrid islanding detection technique using average rate of voltage change and real power shift, IEEE Transactions on Power delivery 24 (2009), 764–771.

28.

Garmrudi

, Nafisi

, Fereidouni

and Hashemi

, A Novel Hybrid Islanding Detection Technique Using Rate of Voltage Change and Capacitor Tap Switching, Electric Power Components and Systems 40 (2012), 1149–1159.

29.

Dhar

and Dash

, Harmonic Profile Injection-Based Hybrid Active Islanding Detection Technique for PV-VSC-Based Microgrid System, IEEE Transactions on Sustainable Energy 7 (2016), 1473–1481.

30.

Shen

J-M.

, Jou

H-L.

, Wu

J-C.

and Wu

K-D.

, Hybrid photovoltaic generation system with novel islanding detection method, Electric Power Systems Research 106 (2014), 101–108.

31.

Rostami

, Abdi

, Olamaei

, Naderi

and Moradi

, A new hybrid islanding detection technique using rate of change of frequency and real power with capacitor switching, 2016.

32.

Pigazo

, Moreno

V.M.

, Liserre

and Dell’Aquila

, Wavelet-based islanding detection algorithm for single-phase photovoltaic (pv) distributed generation systems. Industrial Electronics, 2007 ISIE 2007 IEEE International Symposium on: IEEE; 2007. pp. 2409–2413.

33.

Menon

and Nehrir

M.H.

, A hybrid islanding detection technique using voltage unbalance and frequency set point, IEEE Transactions on Power Systems 22 (2007), 442–448.

34.

Cui

, El-Arroudi

and Joos

, Islanding Detection of Hybrid Distributed Generation under Reduced Non-detection Zone, IEEE Transactions on Smart Grid. 2017; pp. 1.

35.

Raza

, Mokhlis

, Arof

, Naidu

, Laghari

J.A.

and Khairuddin

A.S.M.

, Minimum-features-based ANN-PSO approach for islanding detection in distribution system, IET Renewable Power Generation 10 (2016), 1255–1263.

36.

Ghazi

and Lotfi

, A new hybrid intelligent based approach to islanding detection in distributed generation. Universities Power Engineering Conference (UPEC), 2010 45th International: IEEE; 2010. pp. 1–5.

37.

Bitaraf

, Sheikholeslamzadeh

, Ranjbar

A.M.

and Mozafari

, Neuro-fuzzy islanding detection in distributed generation. Innovative Smart Grid Technologies-Asia (ISGT Asia), 2012 IEEE: IEEE; 2012, pp. 1–5.

38.

Samantaray

, Babu

B.C.

and Dash

, Probabilistic neural network based islanding detection in distributed generation, Electric Power Components and Systems 39 (2011), 191–203.

39.

Shayeghi

and Sobhani

, Zero NDZ assessment for anti-islanding protection using wavelet analysis and neuro-fuzzy system in inverter based distributed generation, Energy Conversion and Management 79 (2014), 616–625.

40.

Zhou

, Cao

, Li

, Cao

, Chen

, Li

, et al., Hybrid islanding detection method based on decision tree and positive feedback for distributed generations, IET Generation, Transmission & Distribution 9 (2015), 1819–1825.

41.

Kong

, Xu

, Yan

, Chen

, Yang

and Han

, Deep learning hybrid method for islanding detection in distributed generation, Applied Energy. 2017.

42.

Kermany

S.D.

, Joorabian

, Deilami

and Masoum

M.A.

, Hybrid Islanding Detection in Microgrid with Multiple Connection Points to Smart Grids using Fuzzy-Neural Network, IEEE Transactions on Power Systems. 2016.

43.

Sareen

, Bhalja

B.R.

and Maheshwari

R.P.

, A Hybrid Multi-feature based Islanding Detection Technique for Grid Connected Distributed Generation, International Journal of Emerging Electric Power Systems (2017), 18.

44.

, Kolwalkar

, Zhang

, Du

and Walling

, Evaluation of anti-islanding schemes based on nondetection zone concept, IEEE Transactions on Power Electronics 19 (2004), 1171–1176.