ReviewBridge the gap: signal processing for power quality applications
Introduction
With the increasing amount of measurement data from power quality monitors, it is desirable that analysis, characterization, classification and compression can be performed automatically. Further, it is desirable to find out the cause of each disturbance, for example, whether a voltage dip is caused by a fault or by some other reason such as motor starting, transformer energizing. Motivated by this, this paper gives an overview of several signal processing techniques that are particularly attractive for these power quality applications. We should point out that, although these signal processing techniques are well known to the signal processing society, the importance lies in the successful applications of these techniques to the addressed power quality problems, and in providing new insight to the power quality data analysis.
Undoubtedly, there is an urgent need for new automatic characterization methods and tools for optimal use of power quality measurements [1], [2], [3]. Another aspect is the need for (real-time) compression of measurement waveform data [4]. Many techniques have recently been developed for extracting features, characterizing and optimally using power quality measurements, and compressing recorded data. These include signal processing methods for extracting features from the measurements, e.g. by Fourier and wavelet transforms combined with neural networks, fuzzy logic, and pattern recognition methods [5], [6], [7], [8], [9], [10], [11], [12]. For classification and characterizing power quality measurements [13], [14] proposed a classification of measurement data using wavelet-based ANN, HMM and fuzzy systems; [15], [16] proposed an event-based expert system by applying event/transient segmentation and rule-based classification of features for different events; [17] proposed a fuzzy expert system. There are also commercial software products for extracting useful information from power quality measurements [9], [18]. For waveform compression of measurement data, the most frequently proposed methods are dyadic wavelet filter-based ones [4], [19], [20], [21]. Each of these methods is shown to be suitable for one or several applications, and is promising in certain aspects in the applications.
Depending on the purposes of disturbance data analysis (e.g. for characterization, feature extraction, modeling) and the associated applications (e.g. diagnostics, classification, protection) different signal processing techniques may be employed. A list of power quality problems will be addressed in the different sections of this paper, with an overview of the state-of-the-art signal processing solutions. They are:
- •
To find the fundamental voltage (or, current) or its harmonics (or sub-harmonics) as a function of time, and to extract information from these components, given the measurements. Section 2 describes the potential methods including time–frequency signal decomposition and harmonic signal modeling.
- •
To detect time instants where voltage (or, current) waveforms undergo sudden changes. Some methods are described in Section 3, including time-scale analysis, and the use of residuals from signal modeling.
- •
To divide measurement data into segments, each containing either a stationary part generated by a single event (the cause of the disturbance), or a transition part between two events. A method is described in Section 4 by using the residuals of Kalman filter models.
- •
To extract features whose space is separable for different types of events. The issue of feature effectiveness is discussed in Section 5 both by examining the separability of the feature space and the empirically determined probability distribution functions of features.
- •
To analyze and understand the causes of voltage transients, Section 7 describes a signal processing method, ESPRIT, to tackle this problem.
- •
To automatically classify events using the extracted features, examples described in Section 6 include both using deterministic and statistical-based approaches. Further, two sample expert systems are briefly described in Section 8 as the applications by combining signal processing techniques with the knowledge of power engineering to solve some complex problems.
- •
To compress the measurement (waveform) data such that the compression is lossless to the application-of-interest (e.g. diagnostic-lossless compression). This will briefly be summarized in Section 9.
- •
To briefly discuss the future research directions in Section 11.
From the examples described below, we shall show that time–frequency, time-scale analysis, model-based methods are useful tools for power system disturbance data analysis and compression. The robustness of each method is largely dependent on the tasks in the applications.
Section snippets
The rms method
Rms is defined for periodic signals, although it is generally used to extract the voltage magnitudes from measurements which are non-periodic. The following formula is typically used for calculating the rms voltage over a multiple of 1/2-cycle (containing N samples) of the power frequency [22]:where Vrms(n) is the voltage as a function of time, sampled at equally spaced points tn=nΔt. The rms values can be computed at each time instant once a new sample is obtained,
Detect transition points
Detecting the time instants where voltage (or, current) waveforms undergo sudden changes (or, transition points) is one of the many essential tasks in disturbance data analysis. For example, one may interested in finding a point-on-the-wave in a recording that might indicate the start of an event, or finding a transition point that might indicate the start of a capacitor switching. Several signal processing examples are given below.
Segmentation of voltage waveforms
For analyzing power system disturbance recordings, it is desirable to partition the data into segments. Segmentation, which is widely used in speech signal processing [29], is found to be very useful in analyzing power quality measurement data as well. In [15], a typical recording of a fault-induced voltage dip is split to three segments: before, during and after the fault. The divisions between the segments correspond to fault initiation and fault clearing. For a dip due to motor starting, the
Select features where feature space is separable for different events
Features can be directly extracted from voltage (or, current) waveforms, however, better features are often extracted from a selected time–frequency, time-scale, or transform domain. Extracting good features that can effectively be used to distinguish different types of events (i.e. the causes of the disturbances) is key to a successful event-based classification system. In this section, we shall consider how to systematically examine the separability of features. Both the feature space and the
Direct approaches
One simple way to classify event-classes, especially when a small amount of data is concerned, is by directly mapping the feature space to a decision space using a discriminant function [30]. The features can be considered as either deterministic or statistical ones. Often, a linear discriminant function (which will find the linear class boundaries) is chosen if there exist training data of known classes (supervised training). For example, a linear discriminant function gi(x)=wtxi+wi0 can be
Analyze and understand the causes of transients
Voltage transients are short duration disturbances to the power system. Voltage transients can be categorized according to their waveform characteristics [33] as being (a) oscillatory transients; (b) impulsive transients.
Oscillatory transients are those whose instantaneous values of voltage rapidly change polarity. The most common causes of oscillatory transients are (a) capacitor energizing; (b) re-strike during capacitor de-energizing; (c) line or cable energizing.
Impulsive transients are
Expert systems for event-based dip classification
In this section, a sample expert system is presented using various combinations of the signal processing techniques described above. The sample expert system is designed to classify the following nine events: (a) Energizing; (b) Non-fault interruption; (c) Fault interruption; (d) Transformer saturation; (e) Induction motor starting; (f) Step change; (g) Transformer saturation followed by protection; (h) Single stage dip due to faults; (i) Multistage dip due to faults.
The expert system, as shown
Compression of power quality data
The large amount of measurement data that calls for automatic analysis techniques, also calls for data compression methods. Data compression may find applications with data storage but also with the transmission of data between computers and a central server. Compression has been an active research area for other digital signals, for example, speech and image signals. Model-based compression, e.g. Linear Predictive Coding (LPC)-based approaches, is mainly used in speech coding where vocal tract
Summary and conclusions
A list of major power quality problems have been addressed where signal processing techniques may offer solutions. We should emphasize that the best suitable signal processing method is dependent on the addressed task, i.e. a method may be the best for one task but may not be suitable for another application task. These are roughly summarized as follows:
- •
For extracting the pseudo fundamental and harmonic contents as functions of time, a range of time–frequency and subband filters may be a good
Discussions: future research directions
For future signal processing research for power quality data applications, we believe that there are several research problems and areas that remain open and are worth of in-depth investigation. Some of these problems are briefly listed below.
- •
Model-based signal processing (including characterization, diagnostics and compression) based on a better understanding of power quality data remains an open research area. Power quality data compression may require more understanding on what is essential
Acknowledgements
Part of this work is supported by Swedish Energy Authority, Elforsk, ABB Corporate research and ABB Automation Products. The authors are thankful to Scottish Power for kindly offering the measurements that were presented in this paper.
References (43)
‘Monitoring power for the future’
Power Eng. J.
(2001)Trends in power quality monitoring’
IEEE Power Eng. Rev.
(2001)Understanding Power Quality Problems: Voltage Sags and Interruptions
(1999)- et al.
‘Power quality disturbance data compression using wavelet transform methods’
IEEE Trans. Power Deliv.
(1997) - et al.
‘Characterization of distribution power quality events with fourier and wavelet transforms’
IEEE Trans. Power Deliv.
(2000) - et al.
‘Power quality detection and classification using wavelet-multiresolution signal decomposition’
IEEE Trans. Power Deliv.
(1999) - et al.
‘Wavelet network-based detection and classification of transients’
IEEE Trans. Instrum. Measurement
(2001) - et al.
‘Classification of power system disturbances using a fuzzy expert system and a fourier linear combiner’
IEEE Trans. Power Deliv.
(2000) - et al.
‘Power quality data analysis: from raw data to knowledge using knowledge discovery approach’
IEEE Power Eng. Soc. Summer Meeting
(2000) - et al.
‘A scalable PQ event identification system’
IEEE Trans. Power Deliv.
(2000)
‘Efficient feature vector extraction for automatic classification of power quality disturbances’
Electr. Lett.
‘FPGA realization of wavelet transform for detection of electric power system disturbances’
IEEE Trans. Power Deliv.
‘Power quality disturbance waveform recognition using wavelet-based neural classifier: theoretical foundation and application’
IEEE Trans. Power Deliv.
‘Power disturbance classifier using a rule-based method and wavelet packet-based hidden Markov model’
IEEE Trans. Power Deliv.
‘Expert system for classification and analysis of power system events’
IEEE Trans. Power Deliv.
‘A novel software implementation concept for power quality study’
IEEE Trans. Power Deliv.
‘Wavelet for the analysis and compression of power system disturbances’
IEEE Trans. Power Deliv.
‘Wavelet-based compression of power disturbances using the minimum description length criterion’
IEEE Power Eng. Summer Meeting
Cited by (68)
A wavelet-based hybrid islanding detection system applied for distributed generators interconnected to AC microgrids
2020, International Journal of Electrical Power and Energy SystemsCitation Excerpt :The combination of islanding detection methods (i.e., passive, active, and communication-based techniques) based on the wavelet transform associated Neural-fuzzy networks determine the proposed PoI [24]. Advanced signal analysis, like a short-time discrete Fourier transform (STFT) [31] and discrete wavelet transform (DWT) [6], can mitigate the NDZ of the PDMs. Despite the STFT employs the time–frequency window to localize the transients, since it uses a fixed window size, the STFT can not operate with multiresolution [32].
Real-time system for automatic detection and classification of single and multiple power quality disturbances
2018, Measurement: Journal of the International Measurement ConfederationEfficient signal processing technique for information extraction and its applications in power systems
2016, Electric Power Systems ResearchCitation Excerpt :However, as stated in [17], by observing that signal processing techniques are mainly designed for speech and image signals, it is certain that a beginning challenge on the development of signal processing in power systems has emerged since new techniques and paradigms can impact measurements and analysis and also enhance quality requirements [18]. In non-stationary signals, where disturbance and abnormal events take place, the use of the FFT becomes inadequate [19]. The STFT divides the signal into time segments, in which the signal is considered stationary.
A method of verifying the statistical performance of electronic circuits designed to analyze the power quality
2016, Measurement: Journal of the International Measurement ConfederationPower quality disturbance detection and classification using wavelet and RBFNN
2015, Applied Soft Computing JournalCitation Excerpt :Many techniques have been proposed in the literature to detect and classify the events envelope. Traditionally, probabilistic approach has been used for time varying signals in a power quality analysis assuming that the power line disturbance components vary too slowly to affect the accuracy of analytical process [1–3]. Another work suggested a combination of spectral method with probabilistic approach, which referred as evolutionary spectrum [4].
A Review on Intelligent Detection and Classification of Power Quality Disturbances: Trends,Methodologies, and Prospects
2023, CMES - Computer Modeling in Engineering and Sciences