An intelligent algorithm for autorecognition of power system faults using superlets

https://doi.org/10.1016/j.segan.2021.100450Get rights and content

Abstract

A time–frequency resolution technique based recognition of power system faults is proposed in the present work. Superlet transformation, a modified and super-resolution form of the wavelet transform, has been used to recognize the type of power system faults occurring in an interconnected power system. The Superlets were tested for sample power system disturbances and found advantageous. An IEEE-9 Bus system has been used to implement the proposed technique wherein it was observed that unique signatures were obtained for each type of fault. Further, to automatically recognize the type of power system faults, a Support Vector Machine (SVM) classifier has been introduced. The SVM is provided with the inputs from the extracted parameters of the proposed Superlets technique. It was found that the proposed methodology of Superlets with SVM has excelled in comparison with other techniques and provided 100% accuracy in the classification of all the events considered.

Introduction

Identification and location of transmission line faults is always a challenge to the power system researchers. The effect of power system faults varies from voltage sags to long time power interruptions depending on the type and location where it has occurred [1]. Utility providers are always concerned with such unforeseen faults and their consecutive effects [2]. The solutions for detecting and localizing such power system transients are still a challenging task to the power system researchers and those R&D organizations which develop devices with modern era algorithms. In this context, extensive research has been reported in the literature using both parametric and non-parametric methods.

Parametric methods are those which are based on statistical theories viz. MUSIC, ARMA, ESPIRIT, Prony analysis, and others. The researchers used these extensively for power system disturbances identification and are still in use as proposed by Javad Khazaei et al.  [3] for real-world PMU data and Sjur Foyen et al. for real-time disturbance measurement using Prony Analysis [4]. On the other side, non-parametric methods, which are signal processing techniques, have taken centre stage in application to power system disturbances. The application of signal processing in power system disturbances analysis has provided a platform for visual monitoring through unique signatures and automatic detection. Extensive research has been done to date in this area, and it is continuously evolving due to its ability to provide faster predictions with high accuracy. In the last five (05) years, the application of time–frequency resolution (TFR) digital signal processing (DSP) techniques has become more prominent as they provide a wide range of unique visual signatures along with highly accurate results. As the research never stops, more new signal processing methods have been developed for other applications, and those are well applied in power systems.

Exploring the literature, Shweta Na et al. have proposed Smoothed Pseudo Wigner–Ville distribution followed by Hilbert transform to identify the type of faults in an IEEE-39 Bus system [5]. The article has reported the simultaneous occurrence of faults also. However, the visual signatures provided in the article were not carrying the information of time–frequency for the types of faults considered. Also, multiple lengths were not considered for a single type of fault. Leandro et al. had proposed smart signal processing techniques for power grids. An IEEE-13 Bus system was considered for obtaining the results, where the wavelet transform was used as a feature extraction tool [6]. A similar wavelet approach was proposed by Sundaravaradan et al. except that it was implemented on a lab-scaled model [7]. The results were accurate, but the system considered was a unidirectional single bus system. The wavelet-based approach was also proposed by Prakash K Ray et al. [8], Nantian Huang et al. [9], T. C. Srinivasa Rao et al. [10] and S. K. Mishra et al. [11] to identify the power system faults. However, the authors have used the indices derived from the modulated wavelets’ energy and have commonly not showcased wavelets’ time–frequency​ properties.

A modified wavelet transform, i.e., Stockwell transform, was also proposed by many researchers to identify power system faults. Nabamita Roy et al. had proposed s-transform with backpropagation neural network to identify power system faults [12]. In-depth results were presented with high accuracy; however, the results were based on only a 2-Bus system, thus restricting the application. Application of sparse S-transform was proposed by Loknath et al. for the location of transmission line faults with UPFC, where the authors have reported that the proposed method is fast compared to ST [13]. Hao Wu et al. have proposed parameters defined on S-transform energy entropy at eight different frequencies to identify power system faults on a T-connection transmission line [14]. However, the accuracy and fastness of the algorithm are not compared with any of the previous methods.

Few researchers have also reviewed the work done on the identification of power system faults using signal processing. Most modern TFR techniques comprising wavelet transform, S-transform, Hilbert transform, and Gabor–Wigner transform, were also discussed in the review articles. As per Manohar Mishra, the wavelet transform experiences spectral leakage and increased computational burden due to decomposition while analysing the power system disturbances [15]. In the case of ST, the direct relation of the Gaussian window width to the central frequency causes harmonic measurement error in real-time applications. Gabor–Wigner transform is having cross-interference issues for applications to power system disturbances, and Hibert–Huang transform requires the right selection of IMFs and end effects of EMD. The facts regarding WT and ST were also mentioned by Ali Raza et al. along with a few other techniques [16].

Conclusively, it is learned that though abundant literature is available about DSP methodologies for power system faults, a new approach with more visual uniqueness, fastness and accuracy is always open for research. Owing to such needs and unexplored areas in the literature, as discussed above, a novel methodology has been proposed in this article. The Superlets transform (SLT) proposed by Vasile V. Moca et al. in 2019 [17] has been used to identify and locate the power system faults in an interconnected power system. This method is used for extracting unique features from various recorded current waveforms and then fed to an artificial intelligence technique for identification.

The major contribution of the proposed methodology are as follows:

  • A high-resolution Superlet technique has been proposed to identify power system faults for the first time in this field of application.

  • Since the power system faults involve high-frequency disturbances for very less time, the proposed high-resolution Superlet technique led to quicker classification and early corrective actions.

  • A unique area under the Gaussian curve method has been proposed, which involves very few steps to calculate, thus reducing the computational burden required for real-time applications.

  • All types of faults have been identified with an accuracy of 100% using Medium Gaussian Kernel of the SVM technique using only two extracted features, i.e., Skewness of Voice (VoS) and Skewness of Local (LoS) as inputs.

  • Using only one extracted feature, i.e., Distance between Fault and No-Fault skewness (DoS), the location of all types of faults is done with a minimum accuracy of 99%, i.e., as a whole only three indices are used to the location and classify the type of fault.

In this paper, the SLT methodology is explained in Section 2, followed by testing and comparing the proposed method for power system disturbances in Section 3. Section 4 is included to illustrate the simulated model considered under the study. The parameter extraction and identification methodology are explained in Section 5, and Section 6 provides the results and discussions. Finally, the conclusions are presented in Section 7.

Section snippets

Superlets Transform (SLT)

The basic idea of superlets transform proposed by Vasile V. Moca et al. in 2019 is to use multiple wavelets to produce a better and less leaky resolution in the time–frequency domain. The drawback of wavelets that an increase in the wavelet bandwidth, i.e., the number of cycles, increases the frequency resolution but at the cost of low temporal resolution. Simultaneously, a lesser number of cycles leads to higher temporal resolution but compromising the frequency resolution. A combination of

Testing of superlets

The SLT is as described in Section 2, is advantageous in terms of maintaining both temporal resolution and frequency resolution. However, the technique’s suitability to power systems disturbances needs to be checked, which is demonstrated in this section. Since power system disturbances comprise of various range of frequencies, the ASLT is used for verification. For comparison, Stockwell transform has been used, as it is a modified, advanced wavelet transform and is reported to be better than

Simulation model

All the waveforms for identifying power system faults are obtained through the IEEE-9 Bus system’s simulated output. It was ensured that all the parameters are in per unit (PU) as defined by the IEEE working group. Both types of faults, i.e., symmetrical and unsymmetrical faults, have been simulated. The fault inception length and location have been moved to obtain waveforms at all three recording stations, i.e., at Bus no. 4, 7 & 9 in the IEEE-9 Bus system as shown in Fig. 4. The faults were

Parameter extraction

Since the response of the ASLT is dynamic and unique for each type of event, the ASLT matrix values have been used for the parameter extraction. The properties of the ASLT matrix is shown below in Fig. 5. The ASLT matrix contains the responses at various times and frequencies, known as locals and voices. A local is the response obtained when the time is constant with variable frequency, and a voice is a variation in response to the change in time at a constant frequency. Both these properties

Results and discussion

The waveforms recorded at Bus 4, Bus 7, and Bus 9 for the faults occurring on the transmission line between Bus 4–5, Bus 7–8, and Bus 9–6 are shown in Fig. 7. The faults have been simulated in the range of 20 Km to 80 Km for a transmission line of 100 Km for an interval of 0.06 s to 0.12 s, i.e., 3 cycles. All types of faults viz three-phase (3Φ), two-phase (2Φ), two-phase ground (2ΦG), and single-phase ground (1ΦG), are considered for simulation and identification. The ASLT for faults at

Conclusions

An adaptive superlet transform (SLT) methodology has been proposed to identify power system faults in an IEEE-9 Bus interconnected power system. The proposed ASLT methodology has been compared with one of the most used time–frequency resolution techniques, i.e., the s-transform, by testing major power quality disturbances. It was found that the ASLT has outperformed ST in terms of both time and frequency resolution. It was further observed that for signals of smaller intervals, an order of 5:20

CRediT authorship contribution statement

Pullabhatla Srikanth: Conceptualization, Data curation, Writing - original draft, Methodology. Chiranjib Koley: Supervision, Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Pullabhatla Srikanth had graduated in Electrical and Electronics Engineering from M. V. G. R. College of Engineering (JNTU Kakinada) in the year 2009. He had completed his post-graduation from the National Institute of Technology (NIT), Hamirpur, in power systems specialization in the year 2011. His research work has been published in various SCI, SCOPUS international journals of repute, and National and International Conferences of IEEE, IET, SPRINGER. He has worked extensively in identifying

References (19)

  • KhazaeiJ. et al.

    Distributed prony analysis for real-world PMU data

    Electr. Power Syst. Res.

    (2016)
  • SankaranC.

    Power Quality

    (2002)
  • SinghB. et al.

    Power Quality Problems and Mitigation Techniques

    (2015)
  • S. Foyen, M.-E. Kvammen, O.B. Fosso, Pronys method as a tool for power system identification in Smart Grids, in: 2018...
  • NaS. et al.

    Detecting instant of multiple faults on the transmission line and its types using time–frequency analysis

    IET Gener. Transm. Distrib.

    (2019)
  • SilvaL.R.M. et al.

    Smart signal processing for an evolving electric grid

    EURASIP J. Adv. Signal Process.

    (2015)
  • AnanthanS.N. et al.

    Real-time fault analysis of transmission lines using wavelet multi-resolution analysis-based frequency-domain approach

    IET Sci. Meas. Technol.

    (2016)
  • RayPrakash

    Detection of faults in a power system using wavelet transform and independent component analysis

    Comput. Commun. Electr. Technol.

    (2017)
  • HuangNantian

    Short-circuit fault detection and classification using empirical wavelet transform and local energy for electric transmission line

    Sensors

    (2017)
There are more references available in the full text version of this article.

Cited by (8)

  • Research on News Push Mechanism Based on Intelligent Algorithm

    2022, Proceedings - 2022 2nd International Symposium on Artificial Intelligence and its Application on Media, ISAIAM 2022
View all citing articles on Scopus

Pullabhatla Srikanth had graduated in Electrical and Electronics Engineering from M. V. G. R. College of Engineering (JNTU Kakinada) in the year 2009. He had completed his post-graduation from the National Institute of Technology (NIT), Hamirpur, in power systems specialization in the year 2011. His research work has been published in various SCI, SCOPUS international journals of repute, and National and International Conferences of IEEE, IET, SPRINGER. He has worked extensively in identifying and analysing power quality problems using signal processing techniques and artificial intelligence. Currently, he is a Ph.D. scholar at National Institute of Technology, Durgapur and working as Scientist D in Defence Research and Development Organisation, India.

Dr Chiranjib Koley, Professor & HOD, has obtained his B.Tech degree from HIT, Haldia, M.Tech degree from IIT, Delhi, and Ph.D. from Jadavpur University, Kolkata in the year 2000, 2002, and 2007 respectively. He is affiliated with Electrical Engineering, National Institute of Technology, Durgapur. Dr Chiranjib Koley has published numerous national and international peer-reviewed journals and presented scientific papers worldwide. Because of the active association with different societies and academies and the contributions, Dr Chiranjib Koley has been recognized by the subject experts around the world. Various reputed awards appreciate his contributions. Dr Chiranjib Koley’s clinical and scientific research interests include Signal Processing, Pattern Recognition, Soft Computing, Process Control, High Voltage Engineering, and Instrumentation.

View full text