Intelligent detection of unstable power swing for correct distance relay operation using S-transform and neural networks

doi:10.1016/j.eswa.2011.05.050

Expert Systems with Applications

Volume 38, Issue 12, November–December 2011, Pages 14969-14975

https://doi.org/10.1016/j.eswa.2011.05.050 Get rights and content

Abstract

The conventional power swing schemes used in distance relay operation are not fast enough to detect and distinguish a fault, stable swing and unstable swing and this may lead to unintended tripping of protection devices. Therefore, there is a need for fast detection of unstable swings so as to improve the reliability of distance relay operation. This paper presents an intelligent approach for detecting unstable swings during distance relay operation using the S-transform signal processing technique and artificial neural networks. To illustrate the effectiveness of the proposed approach, simulations were carried out on the IEEE 39 bus test system using the PSS/E software. Test results showed that the proposed approach using S-transform, multi layer perceptron network and probabilistic neural network can accurately detect and classify fault, stable swing, unstable swing, fault clearance and post fault events for correct distance relay operation.

Highlights

► Power quality monitor placement. ► Particle swarm optimization. ► Artificial immune system.

Introduction

Power oscillation is inherent to power systems which may result from any event such as line switching, short circuit faults, generator tripping or load shedding. During power oscillation, apparent impedance, Z_a at relay location may decrease and enter the relay tripping zone. In this situation, the relay needs to make a proper decision either to activate the tripping signals or to block the tripping signals. (Horowitz & Padke, 2008) and (Jonsson, 2002) have classified the power oscillation into two types, namely, stable swing and unstable swing. During stable swing, tripping actions should be avoided at all costs. However, in case of unstable swing, the tripping signals need to be activated at certain locations, generally at the transfer lines. The tripping actions are essentially important in order to separate the unstable part from the entire power system. The current distance protection utilizes out of step (OOS) detector to distinguish a fault, stable swing and unstable swing. However, many findings have shown that these techniques fail to perform during fast power swing (Abdelaziz et al., 1998, Hossam, 1998, Song et al., 1998). There are a few methods that have been developed in order to improve the reliability of such protection schemes.

The use of artificial neural networks (ANN) for unstable swing detection has been explored (Abdelaziz et al., 1998). However, the application of ANN for this detection was only appropriate for generator side. Song et al., 1998, Hossam, 1998 applied fuzzy theory to distinguish between stable and unstable swings of generators. Nevertheless, this approach is not practical for distance relay protection in transmission lines because it requires features from generator parameters to execute the fuzzy logic operation. Rebizant and Feser (2001) applied two artificial intelligent techniques, namely, ANN and hybrid ANN-fuzzy based protection techniques. Both techniques show very promising results in detecting stable and unstable swings of generators. Again, this technique only focuses on generator protection scheme rather than transmission protection.

As highlighted above, most of the previous studies on the unstable swing detection focuses on the generator operation during power oscillation. However, due the stability and reliability concerns, the development of unstable swing detection technique at transmission network also needs to be enhanced further. Wavelets have been used to extract the features during unstable swing and fault conditions (Brahma, 2007). Wavelet is an advanced signal processing techniques that is widely applied in power system studies. The different levels of wavelets are extracted to identify the features of fault and power swing conditions. Wavelets have been successfully applied for distinguishing the features of unstable swing and fault but it is unable to distinguish between stable and unstable swings (Brahma, 2007). To overcome this problem, a novel algorithm using frequency deviation of voltage was introduced for detecting unstable swing (So, Heo, Kim, Aggarwal, & Song, 2007). The technique is applicable for detecting unstable power swings on transmission lines but however, it is a bit laborious as it needs an additional detector for detecting unstable swing.

To address the need for fast detection of unstable swings so as to improve the reliability of distance relay operation, a new scheme for detecting a fault, stable swing and unstable swing at transmission lines is proposed by using the S-transform and artificial neural networks. The S-transform is used to extract features of signals obtained during fault, stable swing and unstable swing whereas artificial neural networks based on multi layer perceptron network and probabilistic neural network are used to classify either a fault, stable swing or unstable swing for correct distance relay operation as shown in Fig. 1. During a stable swing, it is compulsory to block the tripping signals, while for unstable swing the tripping signals has to be triggered to the associated breaker for isolation purposes.

In the proposed scheme, the input data comprising of bus voltage, active power and reactive power are processed to determine the derivative of bus voltage, and the S-transform features of the voltage and active power for fault, stable swing and unstable swing conditions.

Section snippets

S-Transform theory

S-transform is a time–frequency representation known for its local spectral phase properties. A key feature of the S-transform is its accurate time–frequency (amplitude and phase) domain by employing a moving and scalable localizing Gaussian window (Faisal and Mohamed, 2009, Venkatesh et al., 2008). The basis function for the S-transform is the Gaussian modulated cosine wave whose width varies inversely with frequency. The S-transform of a signal, h(t), is defined by a general equation given as,

Artificial neural network theory

Artificial neural network (ANN) has been utilized for classifying fault, stable swing and unstable swing to assist the correct distance relay operation. In this study, two different ANNs, namely, multi layer perceptron neural network and probabilistic neural network (PNN) have been developed for the same purpose.

Feature selection

Proper selection of input features is an important step before implementing the ANN. The input features of MLPNN and PNN are selected by considering the derivative of bus voltage, the features of bus voltage and bus active power processed by the S-transform. The mathematical formulation of the input features are described accordingly,

Feature 1, F1, is given by, $F 1 = ρ + 1, if \frac{Δ V_{bus}}{Δ T} > 0$ where ρ ∈ {1, 2, 3, …} $F 1 = 0, if, \frac{Δ V_{bus}}{Δ T} ⩽ 0$ Feature 2, F2, is derived from the S-transform and is given by, $ς 2 = |\sum_{m = 0}^{N - 1} V_{bus} [\frac{m + 500}{NT}] Φ|$

Test results

Power swing simulations were carried out to generate training data for the test system consisting of 10 generators, 18 loads and 36 lines as shown in Fig. 6. A fault with duration 350 ms are triggered at different locations of the test system in order to create different cases of stable and unstable swings. Fig. 7, Fig. 8 show the examples of the time domain simulations illustrating cases of stable and unstable swings.

In the time domain simulations, a three phase fault is created in the middle

Conclusion

A new scheme to detect and distinguish unstable swing, stable swing, fault, fault clearance and post fault for correct distance relay operation has been presented. The proposed scheme uses the S-transform for processing the input signals while the PNN is used to automate the event detection process. The MLPNN was also developed to compare its performance with the PNN. The performance of the proposed scheme is investigated through simulation studies for different system parameters and conditions

Acknowledgment

The authors gratefully acknowledge the Universiti Teknologi MARA for financial support in terms of scholarship and Universiti Kebangsaan Malaysia for financial support on the project.

References (10)

A.Y. Abdelaziz et al.
Adaptive protection strategies for detecting power system out-of-step conditions using neural networks
IEEE Proceedings Generation, Transmission and Distribution
(1998)
S.M. Brahma
Distance relay with out-of-step blocking function using wavelet transform
IEEE Transactions on Power Delivery
(2007)
Faisal, M. F., Mohamed, A. (2009). Identification of sources of voltage sags in the Malaysian distribution networks...
Stanley H. Horowitz et al.
Power System Relaying
(2008)
E.A.T. Hossam
Predictive out-of-step relaying using fuzzy rule-based classification
Electr. Power Syst. Res.
(1998)

There are more references available in the full text version of this article.

Cited by (28)

Statistical approach for detection of fault and stable and unstable power swings based on signal energy
2023, International Journal of Electrical Power and Energy Systems
Citation Excerpt :
The method properly operates for high Rf up to 170 Ω, but it does not detect the type of PS. In [13], the fault, SPS, and UPS are detected using probabilistic neural networks and S transforms, while the fault during PS is not investigated. Intelligent methods are accurate but are challenged by the need to teach different scenarios [10].
Distance relays, as the main protection of the transmission lines, are prone to maloperation during power swings. Therefore, the detection of faults and stable and unstable power swings is vital in the proper operation of distance protection. The existence of a method that detects all of them with the same input and calculation can considerably reduce the computational burden of relays and accelerate the performance speed. Therefore, this paper proposes a new statistical approach for the detection of these disturbances based on the voltage energy signal as input. Three indices, including the proposed shape index (PSI) based on the kurtosis and skewness, the proposed dispersion index (PDI) based on the coefficient of variation, and the swing detection index (SDI), are presented to achieve this aim. The method is unaffected by current transformer saturation due to using only voltage signals. Moreover, the method has dynamic thresholds, whose equation coefficients are constant for any system with any voltage level. The method is tested on WSCC 9-bus and New England IEEE 39-bus test systems, considering challenging situations such as noise, high resistance faults, and different swing frequencies. The results of all cases of both test systems show that the method has 100% accuracy in the detection of faults with an average detection time of 0.69 ms in noise-free conditions, while, in noisy conditions, the fault detection accuracy and the average detection time are about 99.37%, and 1.75 ms, respectively. Furthermore, the detection accuracy of stable and unstable power swings is 100% with or without noise.
A new technique for detection and classification of faults during power swing
2019, Electric Power Systems Research
Citation Excerpt :
Phasor based features are extracted from voltage and current signals to train an adaptive neuro-fuzzy interface system (ANFIS) [16] to distinguish faults from power swing. The probabilistic neural network (PNN) is trained by the features extracted using S-transform (ST) to classify fault, stable swing, unstable swing, fault clearance and post fault condition [17] and to discriminate symmetrical fault from power swing [18]. Support vector machine (SVM) is applied to segregate fault, power swing and voltage instability [19] and SVM in combination with multiresolution morphological gradient (MMG) [20] can detect fault quickly during power swing.
The Lissajous figure of voltage and current signals of a transmission line follows a distinct pattern during power swing, although it deviates from its trend when a fault occurs. This feature of Lissajous figure is utilized in this article for detection and classification of faults during power swing. Autoregression technique is applied to obtain the trend of Lissajous figure and the deviation of actual Lissajous figure from this trend is calculated using Euclidean norm. Quarter-cycle moving window sum of Euclidean norm is used to obtain fault detection index for detection of faults during power swing. For identification of faulty phase, fault features are also calculated for three phases and ground using fault signals of only half-cycle duration after fault. The proposed method is tested on a hybrid transmission line consisting of an overhead line and an underground cable, simulated in electromagnetic transient program (EMTP). It can detect faults quickly and classify them with high accuracy irrespective of fault type, fault location, fault resistance, load angle, fault inception time and swing frequency. The method performs satisfactorily even under noisy condition upto 30 dB Gaussian noise. Finally, its performance is compared with one conventional and two novel methods to establish its efficacy.
Pattern-recognition methods for decision-making in protection of transmission lines
2019, Decision Making Applications in Modern Power Systems
Pattern-recognition methods will enable detection, prediction, classification of fault, and decision-making, which are important functioning integrated into the implementation of the protection schemes to develop a smarter transmission system. Electromechanical, electronic, digital, numerical, and today smart relays represent a trend in the field of power system protection and in a special case, protection schemes of transmission lines. In the procedure of pattern recognition, interesting features or useful information can be extracted from power system signals. The discovered features, as an input feature vector, can be applied to decision-making or process of supervision in the protection schemes. The decision-making model contains a representation of the nonlinearity of the mapping between the input vector and output target. Therefore the pattern-recognition method is considered as one of the most important frontiers in complex and nonlinear problems, say power system protections. The proposed chapter will provide a framework for different functions of future technology of smart distance relays. Fault detection, fault classification or phase selection, high-impedance fault detection, symmetrical fault detection during power swing, and power swing detection are the functions developed through pattern-recognition methods. Since the performance of some schemes of the conventional distance relays has some shortcomings, the pattern-recognition methods have the potential to improve the performance of the schemes. They can execute their functionality with cooperation with other routines of the conventional relays. Moreover, the pattern-recognition methods need not exact information on the parameters of power systems to make the final decision, while these methods can work under missing data or noisy conditions. They can also have the capability to adopt against changing the power system topology such as compensated or double-circuit transmission lines. In addition to overcoming some shortcomings of conventional schemes, researchers are interested to propose approaches toward more simplified but more reliable protection schemes. All protection schemes will be examined and evaluated by feasibility, security, dependability, reliability, and speed criteria. To improve the criteria, protection schemes can be addressed by some new approaches such as pattern recognition. This chapter discusses the details of protection schemes based on pattern recognition and also suggests a new trend of setting-less smart protection schemes. Extensions of this idea and giving a new perspective to apply pattern recognition to a future generation of protection relays are the purpose of this chapter.
Progress on power swing blocking schemes and the impact of renewable energy on power swing characteristics: A review
2015, Renewable and Sustainable Energy Reviews
Citation Excerpt :
Finally, the conclusion is drawn in Section 6. Power swing is defined as “large oscillation of power between two areas of a power system” [8]. It can be categorized into two types, stable and unstable swing.
The occurrence of power swing had been reported to threaten the seamless operation of power system. Several cascaded outages and major blackouts in the world have been traced to mal-operation of distance relays due to power swing occurrences. As such, this paper will present a consolidated review on the power swing phenomenon, covering the causes of power swing, its impact to consumers and generators, as well as its detection and relay blocking schemes to prevent the mal-operation of distance relay. Given the wide arrays of power swing detection techniques and blocking schemes available up to date, particular attention is concentrated at the various power swing detection techniques on high voltage power system. In addition, the effect of modern renewable energy generations to the power swing characteristics is also presented in this paper.
A novel sparse approximation-based peak detecting algorithm to differentiate faults from power swings
2023, Electrical Engineering
An overview of power swing detection methods in distance relays and the factors involved
2023, IET Generation, Transmission and Distribution

View all citing articles on Scopus

View full text

Intelligent detection of unstable power swing for correct distance relay operation using S-transform and neural networks

Abstract

Highlights

Introduction

Section snippets

S-Transform theory

Artificial neural network theory

Feature selection

Test results

Conclusion

Acknowledgment

Adaptive protection strategies for detecting power system out-of-step conditions using neural networks

IEEE Proceedings Generation, Transmission and Distribution

Distance relay with out-of-step blocking function using wavelet transform

IEEE Transactions on Power Delivery

Power System Relaying

Predictive out-of-step relaying using fuzzy rule-based classification

Electr. Power Syst. Res.