Elsevier

Measurement

Volume 164, November 2020, 108097
Measurement

Real-time voltage sag detection and classification for power quality diagnostics

https://doi.org/10.1016/j.measurement.2020.108097Get rights and content

Highlights

  • A voltage sag detection and classification method is proposed.

  • Independent component analysis technique is used for detection and segmentation.

  • Higher-order statistics and neural networks are used for classification.

  • It was tested signal windows of 1, 1/2, 1/4 and 1/8 cycle.

  • Good results were achieved in which a detection error rate of 0.86% was achieved.

Abstract

This work proposes an innovative approach to detect, segment and classify voltage sags according to their causes. To detect and segment, Independent Component Analysis is used, with the advantage of being fast and with low computational effort in the operational stage, once it uses only 1/8 cycle of the fundamental component. For classification purposes, Higher-Order Statistics are used for feature extraction and the classifiers are based on Neural Networks and Support Vector Machines. It was tested signal windows of 1, 1/2, 1/4 and 1/8 cycle. For both detection/segmentation design and feature selection, it was used the metaheuristics Teaching-Learning-Based Optimization. Encouraging results were achieved for the simulated signals. In addition, real signals were used to evaluate the detection and segmentation method and good results were achieved in which a detection error rate of 0.86% was achieved.

Introduction

The growing demand for electrical power quality has motivated the increase of power quality (PQ) researches [1], [2], [3]. Many works about PQ perform the classification of PQ disturbances such as voltage sag, harmonics, and transients [4], [5], [6], [7]. On the other hand, other works look for event classification, i.e. identification of the underlying causes of certain disturbance.

Voltages sags, along with interruptions, are the disturbances responsible for 85% of equipment malfunction due to pour PQ, being the main cause of industrial processes stops [8]. Thus, identification of voltage sag causes has gained great attention in works involving PQ [9].

The main causes of voltage sags are: faults, transformer saturation and induction motor starting [10]. Voltage sags caused by faults are the most severe ones. Its magnitude depends on the fault impedance, the distance to the fault point and the system configuration [11]. In cases of induction motor starting, its current can reach up from 6 to 10 times the nominal one, leading to equal drop in all phases and sags with smaller amplitude. The sag magnitude will depend on motor characteristics and the strength of the system [11], [12]. The characteristics of sags caused by transformer saturation are: the three phases are unbalanced, there is no change on phase angle and the presence of even harmonics [13].

In this scenario, automatic detection and classification of voltage sags allow the creation of an event database that can be used for statistical analysis of the Electric Power System (EPS). The fast and online detection of this disturbance is critical for protection systems and to verify compatibility with the ride-through sensitive of equipment nearby. In addition, the classifier can be used to evaluate and optimize the protection system, which often goes into operation unnecessarily, for example, in induction motors starting and transformer saturation, which are mistaken for sags caused by faults, but do not require the operation of the protection system.

Some of the works that deal with sag problems only perform the task of detection/segmentation of signals containing sags while other also perform the classification of these disturbances. Several works have already been developed in this area, such as the ones referred in the next paragraphs.

In [14] the authors proposed a method of detecting sags for applications in dynamic voltage restorers. The method is based on the fundamental amplitude calculation using the d-q components of each phase of the voltage signal in half-cycle windows. The method was efficient for different levels of sags, phase-jump, harmonics, and variation in fundamental frequency.

In [15] an algorithm that uses the Wavelet transform was developed to detect sags in the presence of flicker and harmonics. The method proved to be more robust, more accurate and faster than other three methods used for comparison, based on DQ (direct quadrature), FFT (Fast Fourier Transform) and EPLL (Enhanced Phase Locked Loop). However, the authors did not test the method on real data and did not mention the presence of noise in the simulated data.

In [16], the authors used a causal and anti-causal segmentation method of voltage sags. The method is based on a cumulative sum algorithm, being able to achieve very precise segmentation results. However, due to the anti-causal approach, this technique is only valid in cases of “batch” processing.

In [13], a method was developed to detect and classify the causes of sags using the Hilbert transform and Probabilistic Neural Networks. The authors first use Empirical Mode Decomposition (EMD) to decompose the non-stationary signal into symmetric signals called Intrinsic Mode Functions (IMFs). Then the Hilbert transform is applied to perform the feature extraction. Finally, a probabilistic neural network (PNN) is used to perform the classification of sags caused by faults (short circuits), induction motors starting or transformers energizing. The method is compared with a system based on Wavelet Transform and Multilayer Perceptron (MLP).

In [17] the authors used delayed Legendre wavelet, ant lion optimization algorithm and a classifier ensemble to detect and classify 8 types of sags. In [9], the authors proposed a method based on Multi-Resolution Analysis (MRA) and Support Vector Machines (SVM) to classify sags caused by faults and induction motor starting. In [18], 10 voltage sag characteristic parameters were used in a deep belief network (DBN) for automatic recognition of sag event types, and in [19], S-transform and Extreme Learning Machine were used for the same purpose. Concerning detection and characterization of time-varying non-stationary voltage sags, Hilbert-Huang Transform is considered in [20].

Modern techniques of deep learning also came to be considered in the classification of sags. In [21], the authors identify the source of the sag, as well as the phase in which it occurred using a self-supervised convolutional neural network. Finally, in [22] several feature extraction and classification methods are implemented comparatively for the task of detecting sags.

In this paper, voltage sag detection, segmentation and classification methods are presented. For detection, the Independent Component Analysis (ICA) is implemented with very low computational effort, considering different sag parameters, presence of other disturbances and different noise levels. Two classifiers are tested: Multi-layer perceptron (MLP) [23] and Support Vector Machine (SVM) [24]. The feature extraction is performed by Higher-Order Statistics (HOS) [25] and the feature selection is performed using the metaheuristic TLBO (Teaching-learning-based optimization) [26]. The voltage sags are simulated using a Distributed Generation (DG) network. Synthetic and real data are also used for validation. The novelties of the paper lie in three aspects: (i) the proposition of a new strategy based on ICA and TLBO for sag detection with low computational complexity; (ii) the use of HOS-based features for sag classification; (iii) the use of TLBO to optimize the sag classifiers performance through reducing the number of HOS features used as input in the classifier, avoiding data redundancy and decreasing the classifier computational cost, and optimizing the MLP architecture, choosing the best number of neurons in its hidden layer.

This paper is organized as follows: Section II presents the basics concepts of the algorithms and methods and the proposed methodology; in Section III the achieved results are presented and, finally, in Section IV the conclusions are pointed out.

Section snippets

Methodology

Fig. 1 shows a block diagram that summarizes all stages required to design the proposed method. The first one employs ICA to perform voltage sag detection and segmentation. This stage makes innovative use of an improved version of the method proposed in [19] in terms of both accuracy and computational complexity. The work proposed in [27] executes ICA algorithm for all signal window processed (every iteration), which demands a 4-cycle window to detect and segment voltage sags. Unlike this, the

Detection and segmentation system

As already mentioned, ICA works as a FIR filter in the detection of sags, attenuating low frequencies and amplifying higher frequencies, where the transient part of the sag is found.

Fig. 6, Fig. 7, Fig. 8 show a three-phase signal with voltage sag caused by three-phase fault, induction motor and transformer saturation, respectively, in (a), and the respective method responses in (b). The red line in (b) represents the detection threshold for the respective signal.

From Fig. 8 it is possible to

Conclusion

To design the sag detection and segmentation system, the ICA and TLBO techniques were used. The use of fixed filters projected by ICA and an adaptive threshold were innovative and allowed to achieve good detection and segmentation results for both simulated and real data. The method proved to be robust to the presence of other disturbances (harmonics, flicker and fundamental frequency variation). Two other important advantages are the low computational complexity in the operating phase and the

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.

Acknowledgements

This work was partially supported by CAPES, FAPEMIG, CNPq and INERGE in Brazil.

References (43)

  • P.F. Ribeiro

    Power systems signal processing for smart grids

    (2013)
  • W.L. Rodrigues Junior, F.A.S. Borges, A.F. da S. Veloso, R.de A.L. Rabêlo, J.J.P.C. Rodrigues, Low voltage smart meter...
  • G. Sahu, A. Choubey, Simultaneous power quality disturbances analysis using modified S-transform and evolutionary...
  • Institute of Electrical and Electronics Engineers. IEEE Std 1159-2009. IEEE Recommended Practice for Monitoring...
  • H. Ismail, Z. Zakaria, N. Hamzah, Classification of voltage sag using multi-resolution analysis and support vector...
  • E. Styvaktakis et al.

    Signatures of voltage dips: transformer saturation and multistage dips

    IEEE Trans. Power Delivery

    (2003)
  • A. Foroughi et al.

    Application of Hilbert-Huang transform and support vector machine for detection and classification of voltage sag sources

    Turkish J. Electr. Eng. Comput. Sci.

    (2014)
  • A.K. Sadigh et al.

    Fast and precise voltage sag detection method for dynamic voltage restorer (dvr) application

    Electr. Power Syst. Res.

    (2016)
  • I. Moreno

    Causal and anti-causal segmentation of voltage dips in power distribution networks

    IEEE Lat. Am. Trans.

    (2016)
  • M.K. Saini et al.

    Fractionally delayed Legendre wavelet transform based detection and optimal features based classification of voltage sag causes

    J. Renew. Sustain. Energy

    (2019)
  • Mei, Fei, et al., Online Recognition Method for Voltage sags based on a deep belief network, Energies, vol. 12, no. 1,...
  • Cited by (0)

    View full text