Wavelet transform method for islanding detection of wind turbines

Renewable Energy 38 (2012) 94e106

Contents lists available at ScienceDirect

Renewable Energy journal homepage: www.elsevier.com/locate/renene

Wavelet transform method for islanding detection of wind turbines H. Kazemi Karegar*, B. Sobhani 1 Faculty of Electrical and Computer Engineering, Shahid Beheshti University, Evin Street, Tehran, Iran

a r t i c l e i n f o

a b s t r a c t

Article history: Received 26 January 2011 Accepted 2 July 2011 Available online 11 August 2011

This paper presents a passive islanding detection method for wind turbines. The proposed method is based on voltage measurements and processing of this voltage with a discrete wavelet transform. This method detects the islanding conditions with the analysis of Daubiches wavelets. The studies reported in this paper are based on time-domain simulations using MATLAB, and the feasibility of the proposed method is evaluated with an experimental system. The experimental system parameters are the same as those of the simulated system. The results show that the proposed islanding detection method succeeds in detecting islanding both in the experimental and simulated systems.
2011 Elsevier Ltd. All rights reserved.

Keywords: Distributed generation Islanding Wavelet Wind turbines Wind turbine simulator

1. Introduction The increase of distributed resources in the electric utility systems is indicated due to recent and ongoing technological, social, economical and environmental aspects. Distributed Generation (DG) units have become more competitive against the conventional centralised system by successfully integrating newgeneration technologies and power electronics. Hence, it attracts many customers from industrial, commercial, and residential sectors. DGs generally refer to Distributed Energy Resources (DERs), including photovoltaic, fuel cells, micro turbines, and small wind turbines, and additional equipment [1]. The total global installed wind capacity at the end of 2010 was 430 TWh annually, which is 2.5% of the total global demand. Based on the current growth rates, World Wide Energy Association (WWEA) predicts that, in 2015, a global capacity of 600 GW is possible. By the end of the year 2020, at least 1500 GW can be expected to be installed globally [2]. However, connecting wind turbines to distribution networks produces some problems, such as islanding. Islanding occurs when a DG and its local load become electrically isolated from the utility; meanwhile, the DG produces electrical energy and supplies the local load [3]. Islanding creates many problems in power systems, and the existing standards thus do not permit DGs to be utilised in islanding mode [4]. Some of these reasons are the following [5,6]: * Corresponding author. Tel.: þ98 21 29904170; fax: þ98 21 22431804. E-mail address: [email protected] (H. K. Karegar). 1 Tel.: þ98 21 29904170; fax: þ98 21 22431804. 0960-1481/$ e see front matter
2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.renene.2011.07.002

e e e e

safety hazards for personnel power quality problems for customers load overload conditions of DG out-of-phase recloser connections

Thus, islanding conditions should be detected within less than 2 s [5]. Originally, the methods of islanding detection were divided into two categories: communication and local. Local methods were classified as active and passive techniques, in which active techniques are based on direct interaction with the ongoing power system operation [4]. Some important active techniques are impedance measurement [7], frequency shift and active frequency drift [7], current injection [8], sandia frequency shift and sandia voltage shift [9], and negative phase sequence current injection [10]. Passive techniques are based on measurements and information at the local site. Some techniques are under/over frequency or voltage [7], total harmonic distortions [2], rate of change of frequency [11], non detection zone concept [12], vector surge and phase displacement monitoring [13], rate of change of generator power output [7], and the THD technique [14]. In this paper, a new method based on Discrete Wavelet Transform (DWT) is proposed for islanding detection of wind turbines. The proposed technique, which is suitable for asynchronous DGs, is explained in Section 3. Section 4 explains the simulation and experimental test system used to verify the effectiveness of the proposed technique. Section 5 explores the effectiveness of the proposed technique applied to the simulation and experimental test systems, and Section 6 concludes the paper. The simulation test systems were simulated in MATLAB/SIMULINK using SimPowerSystemBlockSet. The

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

a

c

S

High-pass

Low-pass

A

95

S

H

L

D H

L

b S

H Low-pass

L

High-pass

... 2

cA

2

cD

cD1

cD2

cD3

cA3

Fig. 1. (a) Filtering, (b) Filtering and Downsampling, (c) Decomposition.

simulation and experimental results show that the proposed islanding detection technique works well to discriminate between switching and islanding conditions.

attempt to increase the window size to increase the frequency resolution leads to loss of time information and vice versa [17]. The DWT is “discrete” in terms of the scaling and shifting. The DWT is defined as [18]

2. Wavelet transform

Cðj; kÞ ¼

XX

SðnÞgj;k ðnÞ;

gj;k ˛Z; j˛N; k˛Z

(1)

n˛Z k˛Z

The Wavelet Transform (WT) is a mathematical tool that is similar to a Fourier transform for signal analysis. Wavelet transform can be described with filter bank theory [15], where a wavelet and a scaling function are associated with a low and a high band-pass filter, respectively. Dyadic wavelet filters can be used for Multi Resolution Analysis (MRA) because it gives good time resolution and poor frequency resolution for high frequency components. It is worth noting that good frequency resolution and poor time resolution are normally used for low frequency components. For implementation, the input signal is divided into two components: low and high frequency. The low frequency component is further split into a low and a high frequency component [16]. The WT is more suitable than the Window Fourier Transform (WFT) or the Short Time Fourier Transform (STFT) because the WFT and STFT methods have permanent window widths. The disadvantage of STFT is the trade-off that has to be made between the length of the window and the frequency resolution. Therefore, any

where, gj;k ðnÞ is a time function with finite energy and fast decay called the mother wavelet. j    j gj;k ðnÞ ¼ a0 2 g a0 n  kb0

(2)

The selection of a0 and b0 is dependent on the family of scaled and shifted mother wavelets. To simply, choose a0 ¼ 2 and b0 ¼ 1, and a dyadic-orthonormal wavelet transform is obtained. SðnÞ can be presented as

SðnÞ ¼

XX

dj;k gj;k ðnÞ

(3)

n˛Z k˛Z

The coefficients dj,k are generated by the DWT and called the ‘resemblance indexes’ between the signal and the wavelet. The similarity is strong if the index is large; otherwise, it is slight.

96

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

The DWT of the signal is calculated by passing it through a series of high-pass filters to analyse the high frequencies and through a series of low-pass filters to analyse the low frequencies. Low frequency signals are called approximations, and high frequency signals are called details [19]. The filtering process algorithm is shown in Fig. 1(a). The signal S is passed into a half band low-pass filter and a half band high-pass filter, which give the approximations (A) and details (D). Therefore, the approximations and details can be down sampled by 2. This process gives the wavelet coefficients of the approximations and details as in Fig. 1(b). The mathematical equations are

Start

Voltage Signal

cA½n ¼

X

SðkÞg½2n  k

(4)

SðkÞh½2n  k

(5)

k

Low -pass No

cD½n ¼

X k

3. Proposed algorithm

Decomposition to level 5 of db

d5>3

Yes

Island Fig. 2. Proposed algorithm in order to islanding detection.

A WT is a time-scale presentation of any fixed or non-fixed waveform using basis functions, which, when widened and translated, are called baby wavelets. The advantage of using a wavelet transform is not only the time-scale presentation but also the preservation of both time and frequency information without any resolution reduction, unlike the short-time Fourier transform (STFT), whose resolution is fixed due to the fixed window size. Different wavelet basis functions having different orders can affect the results; therefore, one of the main objectives of this study is to determine the best wavelet basis function to provide accurate results and, simultaneously, have a small number of wavelet coefficients to speed-up the computational process instead. Wavelet db5 is selected, and the decomposition level is set to 5. Here, a new method is implemented for islanding detection based on the wavelet transform of the voltage signal of the local load. In this study, the voltage signal of the load is measured, and then it is passed through the low-pass filter to expurgate the fundamental frequency of the voltage signal. Then using the DWT of this signal and making a comparison with the threshold value, it is determined whether islanding occurred. Fig. 2 shows the proposed algorithm of this method.

Fig. 3. Single line diagram of study system.

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

97

Fig. 4. Single line diagram of implementation system in order to islanding condition detection.

Fig. 6. Implementation system in order to islanding condition detection.

4. Case study

5. Implementation and simulation results

Fig. 3 shows a schematic diagram of a wind turbine unit. The DG unit is a wind turbine induction generator, and a capacitor bank is used to improve the power factor. The local load is a three-phase parallel RL before the circuit breaker (CB), in which “r” denotes the series resistance inductance and Vf indicates the voltage drop across the parallel load. The parallel RL is conventionally adopted as the local load for the evaluation of islanding detection methods when the load inductance is tuned to the system frequency. This system, as shown in Fig. 3, is connected to a Point of Common Coupling (PCC) with a step-up transformer. To obtain the experimental results, a wind turbine simulator, as shown in Fig. 4, was implemented. Fig. 5 and Fig. 6 show the implemented simulator system. The implemented system parameters are given in Table 1. The parallel load inductance is considered infinite. Thus, the parallel load is only a resistance, and hence the unit of “L” is “inf”. Fig. 7 shows the motor saturation curve. In the grid-connected condition, the switches SW1 and SW2 are closed. The islanding condition occurs when SW2 is open. The voltage and frequency of DG should have admissible values in both grid-connected and islanded modes. In the grid-connected mode, the voltage magnitude and frequency of the local load at the PCC are regulated by the grid.

In this study, the simulation is conducted in five scenarios to illustrate the effectiveness of the proposed method. 5.1. Nominal load condition In this case, the load as shown in Fig. 3 is set to the values given in Table 1. The DG is connected to the grid and works in grid-connected mode. At t ¼ 2 s, the CB is opened, and the system enters islanding mode. Fig. 8 shows the dynamic response of the system prior, during and subsequent to the islanding event. Fig. 8(a) shows the instantaneous voltage of phase-a at the PCC. Fig. 8(b) shows the d5 coefficient of the Daubechies wavelet transform. According to this figure, at t ¼ 2.23 s, the d5 coefficient increases from threshold value. Therefore, the proposed method detects the islanding. Fig. 8(c) Table 1 Parameters of the implemented system. Parameters

Value

Induction motors

Sn Vn f PF Rs, Rr Lr, Ls Lm R L C

Local Load Capacitor

2 kVA 400 V 50 Hz 0.78 Lag 2.3541 U 0.01678 H 0.275 H 180 U inf 36.75 mF

500 400

V(v)

300 200 100 0

Fig. 5. Implementation system in order to islanding condition detection.

0

2

4

6 I (A)

8

Fig. 7. Motor and generator saturation curves.

10

12

98

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

Fig. 8. Dynamic response of the simulation system, a) Instantaneous voltage of phase-a, b) d5 coefficient of daubechies, c) RMS PCC voltage, d) PCC voltage frequency.

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

99

Fig. 9. Dynamic response of the experimentally system, a) instantaneous voltage of phase-a, b) d5 coefficient of daubechies, c) RMS PCC voltage.

shows the RMS value of the PCC voltage. Fig. 8(d) shows the frequency of the load and demonstrates that it has no main change prior to the islanding and after the islanding condition. In Fig. 9, the experimental results for the nominal load are depicted. Fig. 9(a) shows the instantaneous voltage of phase-a at the PCC, and Fig. 9(b) shows the d5 coefficient of the Daubechies wavelet transform. As shown, the d5 coefficient is increased from the threshold value at t ¼ 2.14 s, which leads to islanding detection. Fig. 9(c) shows the RMS value of the PCC voltage. The RMS voltage of utility is 385 V before islanding, and after islanding, the voltage decreases to 380 V.

5.2. Mismatch power condition The system shown in Fig. 3 operates in a grid-connected mode. The load absorbs 200 W of real power from the grid and sends 150 var reactive power to the grid and 800 W real power from the DG. The load parameters are R ¼ 144.4 U, L ¼ 3H and C ¼ 43.38 mF. An islanding event occurs at t ¼ 2 s and is detected at t ¼ 2.21 s with the d5 coefficient. The islanding detection time is shorter than that in the previous case study. The voltage magnitude at the PCC and the islanded system frequency change rapidly, but these values do not deviate from their acceptable limits. Fig. 10 shows the results of

100

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

Fig. 10. Dynamic response of the simulation system, a) Instantaneous voltage of phase-a, b) d5 coefficient of daubechies, c) RMS PCC voltage, d) PCC voltage frequency.

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

101

Fig. 11. Dynamic response of the experimentally system, a) Instantaneous voltage of phase-a, b) d5 coefficient of daubechies, c) RMS PCC voltage.

this condition in the simulated system. According to Fig. 10(c), the RMS value of the voltage is increased after islanding, and the frequency of the system, as shown in Fig. 10(d), is decreased. The results for the experimental system are depicted in Fig. 11. Fig. 11(a) shows the instantaneous voltage of phase-a at the PCC. Fig. 11(b) shows the d5 coefficient of the Daubechies wavelet transform. In this case, at t ¼ 2.6 s, islanding occurs and is detected at t ¼ 2.77 s. Fig. 11(c) shows the RMS value of the PCC voltage. The RMS voltage of utility is 385 V before islanding and is increased to 400 V after islanding.

5.3. Motor starting condition The starting of large induction motors may cause a malfunction of the islanding detection algorithm. To study the reliability of the proposed algorithm, a 2 kW induction motor is connected to the PCC via a switch in the non-islanding case. The simulation results of the induction motor starting are shown in Fig. 12, and the experimental results of this condition are shown in Fig. 13. At t ¼ 2 s, the induction motor was started. Then the RMS voltage of the PCC decreased, as shown in Fig. 12(c), but the

102

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

Fig. 12. Dynamic response of the simulation system, a) Instantaneous voltage of phase-a, b) d5 coefficient of daubechies, c) RMS PCC voltage, d) PCC voltage frequency.

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

Fig. 13. Dynamic response of the experimentally system, a) Instantaneous voltage of phase-a, b) d5 coefficient of daubechies, c) RMS PCC voltage.

103

104

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

Fig. 14. Dynamic response of the simulation system, a) Instantaneous voltage of phase-a, b) d5 coefficient of daubechies, c) RMS PCC voltage, d) PCC voltage frequency.

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

105

Fig. 15. Dynamic response of the experimentally system, a) Instantaneous voltage of phase-a, b) d5 coefficient of daubechies, c) RMS PCC voltage.

frequency changed, as shown in Fig. 12(d). Fig. 12(b) shows that the d5 coefficient is not sensitive to motor switching, and its value does not increase within this time. Therefore, the proposed method does not send a trip and works in a reliable mode. In the experimental mode, the switching of the motor occurred at t ¼ 1.2 s, and the voltage of the PCC decreased at this time. The voltage of grid in this study is 395 V and, at the instant of switching, decreased to 375 V. According to Fig. 13 (b), it is obvious that the experimental results and simulation results prove the reliability of the proposed method. 5.4. Capacitor bank switching condition Large capacitor bank switching in distribution power systems initiates disturbances. These disturbances are propagated in the distribution system and have some effects on the proposed method. To test the proposed algorithm, a large 2 kvar capacitor bank was switched at the PCC in the non-islanding case. This switching occurred at t ¼ 2 s. The results for simulation system are

shown in Fig. 14. At the switching time, the RMS voltage and frequency are almost constant. Fig. 14 (b) shows that the d5 coefficient does not change in this condition. The results show that d5 does not have any sensitivity to the switching condition, and the proposed method works perfectly. The experimental results are described in Fig. 15. A 2 kvar capacitor bank was switched at t ¼ 1.1 s. The results confirm the simulation results. 6. Conclusions This paper presents a new method based on wavelet transforms for the islanding detection of wind turbines. The proposed method was simulated and implemented on a wind turbine simulator. The results show the suitable reliability of the proposed method under different load conditions, such as capacitor bank switching and motor starting. Under these conditions, islanding detection is difficult. The method was able to detect the islanding condition of an induction generator type of a wind turbine within less than 0.2 s.

106

H.K. Karegar, B. Sobhani / Renewable Energy 38 (2012) 94e106

Acknowledgement The authors would like to thank to The Vice President for Research and Technology of Shahid Beheshti University for the support received for this work. References [1] Jiayi H, Chuanwen J, Rong X. A review on distributed energy resources and micro-grid. Renew. Sust. Energ Rev. 2008;12:2472e83. [2] http://www.renewableenergyworld.com/rea/news/article/2011/05/worldwind-outlook-down-but-not-out. [3] Jayaweera D, Galloway S, Burt G, McDonald JR. A sampling approach for intentional islanding of distributed generation. IEEE Trans. on Power Syst. 2007;22(2):514e21. [4] Zeineldin HH, Abdel-Galil T, El-Saadany EF, Salama MMA. Islanding detection of grid connected distributed generators using TLS-ESPRIT. Electr. Power Syst. Res. 2007;77:155e62. [5] Xu Wilsun, Mauch Konrad, Martel Sylvain. An assessment of DG islanding detection methods and issues for Canada. CETC-Varennes 2004-074 (TR) 411-INVERT; 2004. 1e6. [6] Swisher R, De Azua CR, Clendenin J. Strong winds on the horizon: wind power comes of age. Proceedings of the IEEE 2001;89(12):1757e64. [7] Chowdhury SP, Chowdhury S, Crossley PA. Islanding protection of active distribution networks with renewable distributed generators: a comprehensive survey. Electr. Power Syst. Res.; December 2009:1e9. [8] Hernández-González G, Iravani R. Current injection for active islanding detection of Electronically-Interfaced distributed resources. IEEE Trans. on Power Deliv. 2006;21(3):1698e705.

[9] John V, Ye Z, Kolwalkar A. Investigation of anti islanding protection of power converter based distributed generators using frequency domain analysis. Trans. on Power Electronic 2004;19(5):1177e83. [10] Karimi H, Yazdani A, Iravani R. Negative-sequence current injection for fast islanding detection of distributed resource unit. IEEE Trans. on Power Deliv. 2008;23(1):493e501. [11] El-Arroudi K, Joós G, Kamwa I, McGillis DT. Intelligent based approach to islanding detection in distributed generation. IEEE Trans. on Power Deliv. 2007;22(2):825e35. [12] Zhihong Y, Kolwalkar A, Zhang Y, Du P, Walling R. Evaluation of anti-islanding schemes based on non detection zone concept. IEEE Trans. of Power Electronic. 2004;19:1171e6. [13] Kunte SR, Gao W. Comparison and review of islanding detection techniques for distributed energy resources. IEEE, Conference; 2008. 1e8. [14] Kazemi Karegar H, Shataee A. Islanding detection of wind farms by THD. IEEE Conf. DRPT2008 Nanjing China, Digital Code: 978-7-900714-13-8/08/
2008 DRPT; April 2008. 6e9. [15] Ukil A, Zivanovíc R. Abrupt change detection in power system fault analysis using adaptive whitening filter and wavelet transform. Electr. Power Syst. Res. 2006;76:815e23. [16] Jurado F, Saenz JR. Comparison between discrete STFT and wavelets for the analysis of power quality events. Electr. Power Syst. Res. 2002;62:183e90. [17] Morsi WG, El-Hawary ME. Novel power quality indices based on wavelet packet transform for non-stationary sinusoidal and non-sinusoidal disturbances. Electr. Power Syst. Res. 2010;80:753e9. [18] Zhao W, Song YH, Chen WR. Improved GPS traveling wave fault locator for power cables by using wavelet analysis. Electri. Power Ener. Syst. 2001;23: 403e11. [19] Zhao W, Song YH, Min Y. Wavelet analysis based scheme for fault detection and classification in underground power cable systems. Electr. Power Syst. Res. 2000;53:23e30.