EEE508 GÜÇ SİSTEMLERİNDE SİNYAL İŞLEME Signal Processing for Power System Applications Triggering, Segmentation and Characterization of the Events (Week-12) Gazi Üniversitesi, Elektrik ve Elektronik Müh. Böl. 2018 Güz Dönemi Prof. Dr. Özgül SALOR-DURNA 1
TRIGGERING AND SEGMENTATION A triggering point is the time instant at which a power quality event starts or ends. Existing methods are simple: Detect changes from waveform or rmd sequences. Partitioning a data sequence into disjoint segments (segmentation) is a necessary preprocessing step before effective data analysis methods can be applied. Although nonstationary signals can be analyzed by using fixedsize block-based methods, it is desirable that the size of data blocks be adaptive according to some given criteria. Segmentation methods aim at optimally choosing variable-size data blocks, each associated with a duration of time within which the system does not change its state or with a transition process between two system states.
TRIGGERING The original distinction between variations and events was that variations are small, slow deviations from an ideal or nominal value whereas events are large, fast deviations. Triggering is the method used to distinguish between small and slow; and large and fast.
TRIGGERING There are two different approaches for the definition of the three types of events. The first approach defines dips, swells, and interruptions as events for which the rms voltage is below 90%, above 110%, and below 10%, respectively. (IEC 61000-4-30). The second approach first determines magnitude and duration of the event and next determines the type of event based on its location in the magnitude duration plane.
IEC 61000-4-30 TRIGGERING
TRIGGERING Consider as an example the event shown in Figure 7.2: a synthetic event due to a single-phase fault that develops into a three-phase fault that leads to the load being disconnected from the supply. The resulting power quality event is, according to IEC 61000-4-30, a dip with a duration T1 + T2 + T3 but at the same time a swell with a duration T1 and an interruption with a duration T3. Note that this event would also pose a problem for IEEE 1159.
TRIGGERING When presenting and interpreting power quality statistics in the form of site and system indices, the presence of these multiple events has to be considered to prevent double counting.
TRIGGERING: TRANSIENTS Triggering typically takes place when the absolute value of the extracted transient exceeds a certain threshold. The difference between the different methods is in the way in which the transient is extracted: The transient is extracted as the output of a high-pass filter. The high-pass filter can be implemented in hardware so that looking for transients does not require any additional computation time. The disadvantage of this method is that high levels of harmonics could lead to false triggering. The transient is extracted as the difference with the previous cycle of the power system frequency. This will require a certain amount of computational time. This method allows for a low threshold setting and thus for a very sensitive detection of transients. However, large frequency deviations may lead to erroneous trips as the buffer length is typically based on the nominal frequency. Triggering takes place when an instantaneous value deviates more than the threshold from the average sine wave. The average sine wave is obtained by averaging over a number of cycles. This method is sensitive to high levels of harmonics and to deviations in the power system frequency.
TRIGGERING: TRANSIENTS The simplest method uses the rate of change of the voltage or current as the triggering criterion: Triggering takes place in that case when OR Wavelets have been used to analyze PQ events and especially transients. The use of wavelets for power quality analysis was originally proposed by Ribeiro, Celio and Samotyj (Future Analysis for Power Quality, EPRI Conference on Power Quality and Applications, 1993).
WAVELETS FOR TRANSIENTS Wavelet theory is the mathematics associated with building a model for a signal with a set of special signals, or small waves, called wavelets. They must be oscillatory and have amplitudes which quickly decay to zero. The required oscillatory condition leads to sinusoids as the building blocks (particularly for electrical power systems). However wavelets do not need to be damped sinusoids. Mathematically speaking, the wavelet transform or decomposition of a function, f(t), with respect to a mother wavelet, h(t), is: Wf ( a, b) 1 = f ( t) h 1 a 2 * t b a dt
WAVELETS FOR TRANSIENTS The Mother Wavelet Scaled and Translated Wavelets
WAVELETS FOR TRANSIENTS The inverse transform creates the original function by summing appropriately weighted, scaled and translated versions of the mother wavelet, as indicated by the following equation. The weights are the wavelet coefficients, Wf(a,b). f ( t) = C h = 1 C h 1 w Wf h( w) ( a, b) 2 dw 1 a t b h a dadb 2 a
WAVELETS FOR TRANSIENTS Wavelets were originally derived to analyze seismic signals in petroleum research. At present they are used in image processing and analysis, and in sound (speech or music) analysis.
WAVELETS FOR TRANSIENTS Original Waveform to be analyzed 2 Wavelet Components Reconstruct function
WAVELETS FOR TRANSIENTS Impulsive Transient Commutation Notches
WAVELETS FOR TRANSIENTS The advantage of using a wavelet-based method over rms methods is that it allows for a much more precise time localization of the event. The wavelet transform can be interpreted as a set of filter banks with increasing bandwidth. In most publications the highest frequency band is used to detect sudden changes in the waveform. This filter has the shortest filter length and thus gives the best time localization. When the change is faster than the time resolution of the signal (i.e., it happens within one time step), the highest frequency band gives the best result. However, in many signals the transition takes several samples. One of the intermediate frequency bands gives a better result in that case.
WAVELETS FOR TRANSIENTS Different authors use different mother wavelets. The two most commonly used ones are the Morlet wavelet and the Daubechies wavelet. The Morlet mother wavelet is a sine wave multiplied by a Gaussian window. The advantage of using a sine wave is that the results can be interpreted in a similar way as the results of the short-term Fourier transform. The Daubechies mother wavelets have a somewhat triangular shape, which makes its interpretation more difficult. 17
TRIGGERING METHODS: Changes in rms waveforms There is a serious limitation of the method; that is, the detected triggering points have rather low accuracy in terms of their time positions. This is due to the nature of the rms where a half-cycle or onecycle window of data is typically used for computing the rms, hence the time resolution of the rms sequence is associated with this window size. 18
TRIGGERING METHODS: High-pass filter + threshold Use a high-pass filter followed by a threshold. This is based on the idea that a high-pass filter can detect the quick changes caused by most underlying system events. The method does not always work well, especially when the changes are not obvious (e.g., a slope change instead of the step change). Further the method is sensitive to noise. 19
TRIGGERING METHODS: Residuals of the models For each sample instant the residual is calculated as the difference between the measured value and the value as predicted by the model. Small residual values indicate an accurate model. For each sample instant the residual is calculated as the difference between the measured value and the value as predicted by the model. Small residual values indicate an accurate model. The start of an event is associated with a sudden change in the system. It is natural to expect that large residuals occur at around the time instant of the underlying system event, The same holds for the end of the event when this is associated with another change in the system. However, for example, a voltage dip due to motor starting will only give one triggering point when using this method. 20
TRIGGERING METHODS: Residuals of the models 21
SEGMENTATION 22
SEGMENTATION Analysis of nonstationary data was performed by first dividing the data into fixed-size blocks before the parameters of the model are estimated. Each block of data is considered as nearly stationary and is characterized by one set of time-invariant model parameters. The main reason for choosing a fixed block size is for simplicity. This is perfectly acceptable if the changes in characteristics are small and slow. For power quality events, however, it is desirable that the block size changes accordingly, so that each block of data (or data segment) can be associated with one underlying cause in the power system. 23
SEGMENTATION An event segment is located between the two adjacent transition segments. A power quality event is characterized by the event segments. The first step of segmentation consists of locating the transition segments. Transitions are caused by system events such as fault initiation, fault clearing by protection operation, induction motor starting, transformer energizing, reclosing following fault clearing, load switching, and capacitor switching, and among others. Due to the fast changes between the power system states and the short duration usually involved in a transition segment, data in a transition segment present different characteristics as compared with those in an event segment. Methods with a higher time resolution may be applied to transition segments, for instance to locate the time instant of an event with high precision. Methods with a high time resolution are often very sensitive, leading to a high risk of false alarm. However, by applying those methods only to the transition segment, the risk of false alarm is significantly 24 reduced.
SEGMENTATION Using model residuals for segmentation: 25
SEGMENTATION 26
SEGMENTATION 27
SEGMENTATION: Examples 28
SEGMENTATION: Examples 29
SEGMENTATION: Voltage Dip Example 30
SEGMENTATION: Voltage Dip Example 31
TRIGGERING AND SEGMENTATION: Existing triggering methods for voltage dips, swells, and interruptions are based on the one-cycle rms voltage. The event is detected when the rms voltage exceeds a threshold: typically 90% of nominal voltage for dips, 10% for interruptions, and 110% for swells. These methods are well defined in an IEC standard. For the detection of transients different methods are in use, including cycle-to-cycle variations and high-pass filters, but no standard method exists. Both triggering and transition segments correspond to a transition between system states. Methods for detecting triggering points and for detecting transition segments are therefore strongly linked with one another. Segmentation should be seen as a more general form of triggering. Whereas triggering results in individual points (the triggering points), segmentation results in intervals (the transition segments). A practical difference between triggering and segmentation is that triggering is always done in real time. 32
CHARACTERIZATION Expert systems are computer systems implemented by methods and techniques for constructing humanmachine systems with specialized problemsolving expertise. The main drawback with expert systems is that the rules of inference must be collected from a human expert and converted to an acceptable form. Disturbance 1: RULE 1: IF 'THD_VOLTAGE'<5% AND RULE 2: IF 'THD_CURRENT'<5% AND RULE 3: 'FUNDAMENTAL_VOLTAGE'<80% THEN DISTURBANCE= PHASE-TO-GROUND FAULT' Disturbance 2: RULE 1: IF 'THD_VOLTAGE'<5% AND RULE 2: IF 'THD_CURRENT'<5% AND RULE 3: 'FUNDAMENTAL_VOLTAGE'<85% AND RULE 4"FUNDAMENTAL _CURRENT>500% THEN DISTURBANCE='MOTOR STARTING