Electrical Power and Energy Systems 30 (2008) 575580

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

Enhancement of islanding-detection of distributed generation systems via wavelet transform-based approaches

Department of Electrical Engineering, Kun Shan University, Tainan 70101, Taiwan Department of Electrical Engineering, Far Eat University, Tainan 70101, Taiwan c Department of Electrical Engineering, National Cheng Kung University, Tainan 70101, Taiwan
b

a r t i c l e

i n f o

a b s t r a c t
In this paper, a wavelet transform-based approach is proposed to detect the occurrence of islanding events in distributed generation systems. Thanks to timefrequency localization capabilities exhibited by wavelet transform, the approach embedded with this transform technique has grasped the appearance of the islanding event in a highly effective manner. Moreover, for those regions which are in need of a better visualization, the proposed approach would serve as an efcient aid such that the mains power disconnection can be better distinguished. To validate the feasibility of this approach, the method has been validated through several scenarios. Test results supported the effectiveness of the method for the application considered. 2008 Elsevier Ltd. All rights reserved.

Article history: Received 27 April 2007 Received in revised form 16 July 2008 Accepted 15 August 2008

Keywords: Distributed generation systems Islanding Wavelet transform

1. Introduction With the rising consciousness of environmental protection as well as the availability of fossil fuels, the installation of distributed generation systems has shed new light as alternative resources in energy supply. In Taiwan, distributed generators are being promoted and installed in parallel with the utility network. It was reported that with the support of such distributed generations, the efciency and stability of supplying power would be signicantly enhanced. Yet, in order to achieve a more secure and economic operation, several technical obstacles are critically required to overcome. This concern arises because the power output of distributed generation systems may not be directly under the control of utility engineers, where the occurrence of islanding is particularly deemed an urgent issue [13]. Fig. 1 depicts a scenario of islanding, where the load-of-interest is severed off from the grid but the system continues to operate because of connected distributed generators. Under this situation, a so-called island is formed, resulting in unexpected consequences that may includes an increased complexity of orderly restoration, a degraded stability of system voltage, and worst of all, a raised risk to related maintenance personnel. In other words, under the scenario of islanding, linecrew members may misjudge the load-side of the line as inactive, where distributed generations are indeed feeding power to loads; hence jeopardizing the life of operators and meanwhile illuminat* Corresponding author. Tel.: +886 6 2757575 32506; fax: +886 6 234 5482. E-mail address: [email protected] (S.-J. Huang). 0142-0615/$ - see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijepes.2008.08.006

ing the importance of a reliable forewarning mechanism to such events. Recently, in order to signal the islanding events, several methods have been suggested [1,413]. Among these published techniques, the supervision of auxiliary contacts of circuit breakers between utility networks and distributed generations was rst considered, where a transfer trip scheme was utilized in order to disconnect distributed generators from the mains supply [7]. Then, once the utility supply is restored, the distributed generators would be resynchronized to the grid. This method was conceptually feasible, yet because its effectiveness is highly dependent on the monitoring performance, it is often hard to implement in real-world applications. Subsequently, several anti-islanding solution approaches were also proposed; which can be largely categorized into two groups: active methods and passive methods [1,8 13]. While active methods examine the operation of a power system in a direct manner, passive methods justify the event based on the system parameters. In active methods, the main theme exists in the design of control circuits such that the required variations can be produced at the outputs of distributed generators. Then, once the loss of grid takes place, this designated bias will accordingly enlarge sufciently to trip the connected relays, notifying the occurrence of the event. On the contrary, when the utility supply is normally operated, the amount of variations will be insufcient to trip the relays, ensuring that there is no event misidentied. Under several circumstances, this active method has won the conrmation; however, the complicated control circuit for the generation of

576

C.-T. Hsieh et al. / Electrical Power and Energy Systems 30 (2008) 575580

Distributed generator

1
where w() is the wavelet basis function, a is the scaling factor, and b is the time shifting factor. When the center and radius of w(t) are individually represented by t* and Dw, the function of has the center of b + at* and the radius of aDw. These parameters can be calculated as follows:

Utility network
:transformer :load

Isolated island

1 jjwjj2 2

tjwtj2 dt
1

2 1 2 3

Fig. 1. Scenario of islanding operation.

Dw
designated bias may offset its merits [46]. As for the passive techniques, they were suggested based on the measurements of system parameters, which may include phase displacement, system impedance, and the change rate of output power. Based on the deviated voltage, current, or frequency following the loss of utility supply, passive methods would justify the islanding through the monitoring of these parameters [12,13]. Yet, without an accurate understanding of parameter variations in the passive method, the possibility of false alarm may largely increase. In the last decades, with the emergence of wavelet transforms, it has been employed in numerous applications with great appreciation [1417]. Based on a family of basis functions, wavelets can be formulated to describe signals in a localized time and frequency format [18,19]. By employing the long windows at low frequencies and short windows at high frequencies, the wavelet transform is capable of comprehending the time and frequency information simultaneously. Hence, the discontinuities and transients in the time-varying signals would be better supervised, thereby motivating the application to enhance the islanding-detection in this study. In this paper, a wavelet transform-based approach is proposed to monitor the parameter variations of interests, where Daubechies wavelet serves as basis. Enhanced by such an approach, it is anticipated that any abrupt change occurred in the acquired signal would be effectively caught, hence increasing the reliability of islanding-detection. Some useful features of this new method are listed below: (1) It helps improve the islanding-detection capability of protective relays. The safety of utility engineers is, meanwhile, better ensured. (2) Because the time and frequency information can be simultaneously observed, the robustness of the method can be better realized for the application considered. (3) With the increased number of installed distributed generators, the proposed method would serve as a potential alternative in addition to existent approaches. (4) The method is easy to program, facilitating the rmware realization of integrated circuit design for the portable detector applications. This paper is organized as follows. Paradigms of the approach are introduced in Section 2. Computation procedures are delineated in Section 3. Numerical tests under different scenarios are simulated and discussed in Section 4. Conclusions are drawn in Section 5. 2. Wavelet formulation Following the emergence of wavelet transform, it has won a high acclaim from several industry applications. The wavelet transform of a signal f(t) can be expressed as:

1 jjwjj2

Z

t t 2 jwtj2 dt

where jjwjj2 is the norm of w() and the time window becomes tw b at aDw ; b at aDw . Note that due to the rapid diminishing feature of wavelet basis function, for those regions outside the time window, they can be neglected. Then, by use of Parsevals identity, the wavelet equation can be represented in the frequency domain

4
b where wx is a window function and bx is the function of f(t) exf pressed in frequency domain. Now, as for the selection of wavelet basis function, the Daubechies wavelet was adopted. We favored this wavelet basis because its formulation was effectively realized as a FPGA chip in our laboratory [20]. Fig. 2 describes a three-band decomposition of wavelet transform process. In the gure, the process starts with a low-pass lter h[n] and a high-pass lter g[n] that decompose the original signal S0[n] into L1[n] and H1[n]. Then, the post-ltered signal L1[n] can be further decomposed into L2[n] and H2[n] through the digital lter h[n] and g[n], respectively. Meanwhile, the signal L1[n] can be also computed from the convolution of S0[n] with h[n] plus a down-sampling with a factor of two, while the signal H1[n] is obtained from the convolution of S0[n] with g[n] plus a down-sampling with a factor of two. By performing this operation, the signal of L1[n] and H1[n] will individually contains lower frequency components and higher frequency ones. It is worth

S0[n] h[n] g[n]

low-pass filter 2

high-pass filter 2

L1[n] h[n]

H1[n] g[n]

low-pass filter 2

high-pass filter 2

downsampling

L2[n]

H2[n]

Fig. 2. Wavelet decomposition.

C.-T. Hsieh et al. / Electrical Power and Energy Systems 30 (2008) 575580

577

mentioning that although a higher band operation is useful to analyze the signal characteristics, the required computation time will grow signicantly as well. Therefore, in consideration of software cost and hardware investment, a three-band decomposition embedded with Daubechies wavelet and a four-coefcient lter was served as the paradigm in this study. The aforementioned decomposition process along with low-frequency Lk and high-frequency components Hk of a signal Sn can be formulated as:

Utility network
CB Load Load
islanding detection CT under/over voltage PT under/over freq. overcurrent

Protective device

Distributed generation

Lk Hk

2N1 X m0

hm S2km1 1m h2N1m S2km1

5 6

Fig. 3. Schematic diagram of the proposed system.

2N1 X m0

For example, when N = 3 and k = 1, 2,. . .,5, the equations can be derived as the following matrix form:

Acquisition of voltage and frequency signals Calculation of voltagedetected magnitude VD Calculation of frequencydetected magnitude FD No

L hS H gS 2 h0 6 60 6 h6 0 6 6 4 h4 h2 2 h5 6 60 6 g6 0 6 6 4 h1 h3 h1 0 0 h2 h0 0 h3 h1 0 h4 h2 h0 h5 h3 h1 0 h4 h2 0 7 07 7 h3 h4 h5 7 7 7 h1 h2 h3 5 0 h0 h1 0 0 0 h1 h0 0 h5 0 h3 h5 0 h2 h4 0 h1 h3 h5 0 0 3

7 8

h5 0 0 0 0 h0 h3 h4 h5 0 0 0 h4 h3 h3 h1 h0 0 h5 h4 h3 h2 0 h0 h3 0 0 h1 0 0 h0 h5 0 0 h4 0 0

0 7 0 7 7 h0 7 7 7 h2 5 h4

3 10

VD > Vth ? Yes FD > Fth ? Yes Islanding detected

Other disturbances

No

where S stands for the original signal, L a column vector of low-frequency components, H a column vector of high-frequency components, h a matrix of low-pass lter coefcients, and g a matrix of high-pass lter coefcients. Next, with wavelet coefcients of Table 1 that was derived from Daubechiess wavelet when N = 2 or N = 3, the matrix computation of (7) and (8) can be performed, thus completing the decomposition task such that the multiple resolution of the signal can be achieved. 3. Computation procedures Fig. 3 delineates the schematic diagram of the proposed system, in which a protective device containing functions of islanding detection, under/over voltage detection, overcurrent detection and under/over frequency detection is delineated. This device can be implemented with an integrated chip such that the signal supervision performance can be upgraded. Computation procedure of this method can be programmed into this protective device. Fig. 4 outlines the ow chart of the computation, where the voltage-detected magnitude and frequency-detected magnitude is individually named as VD and FD. When both of VD and FD are larger than the thresholds of Vth and Fth, the islanding event will be

Fig. 4. Flow chart of the proposed approach.

agged. In the study, both thresholds were set to be 0.1. These settings were made based on several tests as well as discussions with utility engineers. Note that special attentions must be prudently paid on the determination of these thresholds. An inadvertent choice of threshold may downgrade the overall detection performance. 4. Numerical simulations To validate the effectiveness of the method, the proposed method has been applied to examine different scenarios. The simulated signals were programmed and generated using C++ language. For each scenario, the rated voltage and frequency of distributed generation system were assumed to be 110 V and 60 Hz, respectively. The sampling rate is 7680 samples per second and the sampling time lasts 0.3333 s. 4.1. Loss of grid

Table 1 Wavelet coefcients N=2 h0 h1 h2 h3 0.4829629131445341 0.8365163037378077 0.2241438680420134 0.1294095225512603 N=3 h0 h1 h2 h3 h4 h5 0.3326705529500825 0.8068915093110924 0.4598775021184914 0.1350110200102546 0.0854412738820267 0.0352262918857095

4.1.1. Test 1: active power unbalance In this test, a distributed generator is assumed to separate from the grid after the occurrence of a fault, where the event is assumed to take place at 0.05 s. Following this loss of utility, a slight mismatch of active power is observed. In Fig. 5a and b, the waveforms of voltage and frequency is individually depicted. Then, in Fig. 5c and d, detection results are plotted using the proposed method. From these plots, both voltage-detected and frequency-detected

578

C.-T. Hsieh et al. / Electrical Power and Energy Systems 30 (2008) 575580

a
Voltage(pu)

a
Voltage(pu)
0 0.05 0.1 0.15 0.2 0.25 0.3

-1

-1 0 0.05 0.1 0.15 0.2 0.25 0.3

Time (s)

Time(s)

b 60.5
Frequency(Hz)
60

60.5

Frequency(Hz)
0 0.05 0.1 0.15 0.2 0.25 0.3

60

59.5

59.5

Time (s)

0.05

0.1

0.15

0.2

0.25

0.3

Time(s)

c
Voltage-detected magnitude (VD)

1.6 1.2 0.8 0.4 0 0 0.05 0.1 0.15 0.2 0.25 0.3

1.5 1

Voltage-detected magnitude (VD)

0.5 0

Time (s)

0.05

0.1

0.15

0.2

0.25

0.3

Time(s)

d
Frequency-detected magnitude (FD)

0.6 0.4 0.2 0

Frequency-detected magnitude (FD)

0.5 0.4 0.3 0.2 0.1 0 0 0.05 0.1 0.15 0.2 0.25 0.3

Time (s)
Fig. 5. (a) Acquired voltage signal, (b) acquired frequency signal, (c) voltagedetected magnitude and (d) frequency-detected magnitude.

Time(s)
Fig. 6. (a) Acquired voltage signal, (b) acquired frequency signal, (c) voltagedetected magnitude and (d) frequency-detected magnitude.

magnitudes become signicant after 0.05 s, reaching a full agreement with the simulated scenario. This case also implies that because the proposed method is able to present the time and frequency information simultaneously, the occurrence of islanding events can be more accurately identied. 4.1.2. Test 2: reactive power unbalance This case examines the response to a loss of grid. A slight reactive power mismatch following the loss of grid is measured. The event is assumed to occur at 0.05 s. Fig. 6a and b plots the acquired voltage and frequency signals, and Fig. 6c and d delineates voltagedetected and frequency-detected magnitude, respectively. Even though the power mismatch is very limited, the deviation observed at both system voltage and frequency effectively help signal the occurrence of the event, exhibiting the feasibility of the method for this case. 4.2. Immunity of disturbances The disturbance-immunity of the method is one of crucial considerations. Such concerns arise because unexpected disturbance may inuence the reliability of islanding-detection. It is critical

that the method should be able to distinguish the islanding event from unrelated interferences. 4.2.1. Test 1: effects of voltage sag This test investigates the immunity of the method to voltage sag disturbance, where a 15% sag disturbances lasting for three cycles is simulated as Fig. 7a and b depict. Then, by using the proposed approach, the detection results are drawn in Fig. 7c and d. For this case, although the voltage-detected magnitude is signicant, the frequency-detected magnitude is smaller than the threshold. Meanwhile, as the wavelet transform perform the event identication by relying on both time and frequency domain of signals, only the time or frequency disturbance would not trigger the false alarm, indicating that islanding-detection signal is not activated in this case, thus supporting the disturbance-immunity capability of the proposed method. 4.2.2. Test 2: effects of notch disturbance Notch can be deemed a periodic transient, which often occurs as a result of phase-to-phase short circuits in the commutation process of converters. In this test, effects of notch disturbance to the

C.-T. Hsieh et al. / Electrical Power and Energy Systems 30 (2008) 575580

579

a
Voltage(pu)

a
Voltage(pu)

-1

0.05

0.1

0.15

-1

0.05

0.1

0.15

Time(s)

Time(s)

b
Frequency(Hz)

60.5

b
Frequency(Hz)
60

60

59.5

0.05

0.1

0.15

0.05

0.1

0.15

Time(s)

Time(s)

c
Voltage-detected magnitude (VD)

2 1.6 1.2 0.8 0.4 0 0 0.05 0.1 0.15

c
Voltage-detected magnitude (VD)

12 8 4 0

0.05

0.1

0.15

Time(s)

Time(s)

Frequency-detected magnitude (FD)

0.3 0.2 0.1 0 0 0.05 0.1 0.15

Frequency-detected magnitude (FD)

0.4

0.4 0.3 0.2 0.1 0 0 0.05 0.1 0.15

Time(s)
Fig. 7. (a) Acquired voltage signal, (b) acquired frequency signal, (c) voltagedetected magnitude and (d) frequency-detected magnitude.

Time(s)
Fig. 8. (a) Acquired voltage signal, (b) acquired frequency signal, (c) voltagedetected magnitude and (d) frequency-detected magnitude.

islanding detection are assessed. Fig. 8a and b individually draws the waveform of the simulated voltage and frequency signals. The notch disturbances are assumed to start at 0.05 s and stop at 0.1 s. Fig. 8c and d shows the detection outcome of using the proposed method. It is seen that even though the voltage-detected magnitude VD is signicant, the frequency-detected magnitude FD is far below the threshold, revealing that no islanding event is misidentied. This test outcome reaches a full agreement with the assumed scenario. 5. Conclusions Following the increased number and enlarged size of distributed generating units installed in a modern power system, the protection against islanding has become extremely challenging nowadays. In this paper, a detection framework enhanced with wavelet transform is proposed. By case studies with numerical simulations, the proposed approach was veried with feasibility, exibility and robustness. This method is being employed to monitor the islanding event for those distributed generating units installed at a petroleum company located in southern Taiwan. A

eld-programmable gate array (FPGA) chip embedded with the proposed method for a portable detector is also under construction and assembling. Test results will be reported in the near future. Acknowledgements The authors are greatly indebted to Taiwan Power Company for providing their valuable operating experiences. This work was partially supported by the National Science Council, Taiwan, under the contract number NSC92-2213-E-006-085. References
[1] Smith GA, Onions PA, Ineld DG. Predicting islanding operation of grid connected PV inverters. IEE Proc Electr Power Appl 2000;147(1):16. [2] Vachtsevanous G, Kang H. Simulation studies of islanded behavior of gridconnected photovoltaic systems. IEEE Trans Energy Convers 1989;4(2):17783. [3] Zeineldin HH, El-Saadany Ehab F, Salama MMA. Impact of DG interface control on islanding detection and non-detection zones. IEEE Trans Power Delivery 2006;21(3):151523. [4] Lopes LAC, Zhang Y. Islanding detection assessment of multi-inverter systems with active frequency drifting methods. IEEE Trans Power Delivery 2008;23(1):4806.

580

C.-T. Hsieh et al. / Electrical Power and Energy Systems 30 (2008) 575580 [13] Jang S, Kim K. An islanding detection method for distributed generation algorithm using voltage unbalance and total harmonic distortion of current. IEEE Trans Power Delivery 2004;19(2):74552. [14] Daubechies I. Ten lectures on wavelets. Philadelphia, USA: Society for Industrial and Applied Mathematics; 1992. [15] Deng H, Ling H. An efcient wavelet pre-conditioner for iterative solution of three-dimensional electromagnetic integral equations. IEEE Trans Antennas Propagation 2003;51(3):65460. [16] Turkheimer FE, Aston JAD, Banati RB, Riddell C, Cunningham VJ. A linear wavelet lter for parametric imaging with dynamic PET. IEEE Trans Med Imaging 2003;22(3):289301. [17] Lai MJ. On the digital lter associated with Daubechies wavelets. IEEE Trans Signal Process 1995;43(9):22035. [18] Huang SJ, Hsieh CT. Coiet wavelet transform applied to inspect power system disturbances-generated signals. IEEE Trans Aerospace Electron Syst 2002;38(1):20410. [19] Ibrahim WRA, Morcos MM. Articial intelligence and advanced mathematical tools for power quality applications: A survey. IEEE Trans Power Delivery 2002;17(2):66873. [20] Huang SJ, Yang TM, Huang JT. FPGA realization of wavelet transform for detection of electric power system disturbances. IEEE Trans Power Delivery 2002;17(2):38894.

[5] Hernndez-Gonzlez G, Iravani R. Current injection for active islanding detection of electronically-interfaced distributed resources. IEEE Trans Power Delivery 2006;21(3):1698705. [6] Karimi H, Yazdani A, Iravani R. Negative-sequence current injection for fast islanding detection of a distributed resource unit. IEEE Trans Power Electr 2008;23(1):298307. [7] Ropp ME, Begovic M, Rohatgi A. Analysis and performance assessment of the active frequency drift method of islanding prevention. IEEE Trans Energy Convers 1999;14(3):8106. [8] Hung GK, Chang CC, Chen CL. Automatic phase-shift method for islanding detection of grid-connected photovoltaic inverters. IEEE Trans Energy Convers 2003;18(1):16973. [9] Imece AF, Jones RA, Sims TR, Gross CA. An approach for modeling selfcommutated static power converters for photovoltaic islanding studies. IEEE Trans Energy Convers 1989;4(3):397401. [10] Hopewell PD, Jenkins N, Cross AD. Loss-of-mains detection for small generators. IEE Proc Elect Pow Appl 1996;143(3):22530. [11] OKane P, Fox B. Loss of mains detection for embedded generation by system impedance monitoring. In: IEE conference on development in power system protects, Nottingham, UK; March 1997. p. 958. [12] Redfern MA, Usta O, Fielding G. Protection against loss of utility grid supply for a dispersed storage and generation unit. IEEE Trans Power Delivery 1993;8(3): 94854.