R G Alcorn, W C Beattie. The Queen s University of Belfast

Transcription

1 POWER QUALITY ASSESSMENT FROM A WAVE-POWER STATION R G Alcorn, W C Beattie The Queen s University of Belfast SUMMARY A wave-power station produces electricity by converting sea-wave energy into electrical energy. This can be done by various methods, but the method considered in this paper uses the oscillating water column principle. This system uses the oscillating wave motion to produce a reciprocating air flow which is rectified into useful torque by a Wells turbine. The station examined in this paper is the recently commissioned LIMPET station on Islay, a small island off the West coast of Scotland. The LIMPET station has two contra-rotating Wells turbines which spin in the same direction irrespective of the direction of air flow across them. Each of these turbines is coupled to an inverter controlled 25kW 415V 3-phase 6-pole wound-rotor induction machine, giving the station a rated capacity of.5mw. The major challenge of such a station is in producing an acceptable quality of supply. This is due to the oscillating nature of the source as there are power variations with each wave and with each group of waves. This paper will describe the assessment of power quality from the LIMPET station with regard to voltage flicker. The phenomenon of flicker will be described, and a method of producing flicker values from pre-recorded voltage data will be presented. The method shown is the production of a software flickermeter based on the IEC flicker standards. The flicker level is calculated using a statistical method that produces a cumulative distribution function from a flicker level classifier. This method of calculation easily lends itself to implementation in software and can be tested against waveforms given in the IEC standard. Although the flickermeter model produced complies to the IEC standards, it will been shown that it is possible to greatly reduce the sample frequency of the device for wave-power applications without a reduction in accuracy. This is due to the fact that a wave-power station will only generate low frequency voltage fluctuations. This novel method of flickermeter implementation requires extra statistical manipulation in order to comply to the standards and this will be detailed in the paper. Finally, several data sets collected from the LIMPET station will be analysed and the results will be presented. These will show that the station is operating well within acceptable limits of voltage flicker.

2 POWER QUALITY ASSESSMENT FROM A WAVE-POWER STATION R G Alcorn, W C Beattie The Queen s University of Belfast ABSTRACT This paper describes the development and testing of a software based flickermeter used in order to assess the supply quality from the LIMPET (1,2) wave-power station on Islay. It describes the phenomenon of voltage flicker and the effect that a wave-power station has on this quantity. The paper also explains techniques developed in order to improve flickermeter performance when used with pre-recorded data. It will also show that the standard flickermeter sample frequency may be reduced for wave-station applications. Finally the paper will present flicker results from preliminary data collected from the LIMPET station and will show that the device is operating well within acceptable limits. INTRODUCTION The major challenge of a wave-power station is in producing an acceptable quality of supply. This is due to the variable nature of the source as there are power variations with each wave and with each group of waves (3). The difference between calm and rough seas produces long-term variations in the average power output, but this presents few problems. However, the fluctuations caused by individual waves must be controlled and smoothed to provide acceptable power. The LIMPET station is mounted on an exposed western facing coastline on the island of Islay situated off the West coast of Scotland. The station is of shoreline oscillating water column (OWC) design. The incoming wind waves produce an oscillation within the column. This oscillation drives air to and fro across a pair of contra-rotating turbines. Figure 1 LIMPET Schematic The water/air interface provides a gearing ratio from the slow motion of the water column to much higher velocity air motion suitable for driving turbines. The turbines are specialised Wells turbines which spin in the same direction irrespective of the direction of air flow across them. Each of these turbines is coupled to an inverter controlled 25kW 415V 3-phase 6-pole woundrotor induction machine, giving the station a rated capacity of.5mw. A schematic of the plant showing its layout and most important elements is shown in Figure 1. There are regulations which must be observed for the generation of electrical power (4,5). These are quite stringent when connected to an isolated point on the grid. In the case of the LIMPET station, it is located at the end of 1km of 11kV line. This paper discusses the impact of a wave-power station on voltage flicker. It also shows the development of a software flicker tool which may be use to analyse past voltage data. The software flickermeter was developed in such a way that it could be included in system simulation in order to predict flicker performance or it could be used on data previously collected from the station to assess power quality. This paper deals with only previously recorded data from the recently commissioned LIMPET station. At the time of publication, only limited data was available with the station operating below capacity, but this will be updated in the future once a database has been assembled of the station in various operating conditions. FLICKER DEFINITION Voltage flicker is a systematic variation of the voltage envelope or a series of random voltage changes, the magnitude of which does not normally exceed the voltage regulations laid down by the supply authority (4,5). Unlike harmonics, the distorted waveform causing flicker contains frequencies below the fundamental (typically below 25Hz) which makes even a single-step voltage change classify as flicker. The most important impact of these fluctuations is that they cause variations in the light output of various lighting sources. It is the human perceptibility to these fluctuations that form part of the system of defining and calculating the flicker severity. The eye-brain chain is most sensitive at a frequency of 8.8 Hz. A sine modulated voltage wave at this frequency with a modulation amplitude of.25% RMS will correspond to a flicker level of 1. This definition was made by exposing a group of people to a flickering tungsten-filament light at a set modulation frequency. The amplitude of modulation was increased and when 5% of the people noticed flickering, the flicker level

3 was set to one. This was repeated over a range of frequencies in order to obtain a perceptibility curve. The IEC perceptibility curve is shown in Figure 2. Perceptibility (p.u.) Frequency (Hz) Figure 2 - IEC Flicker Perceptibility Curve In most cases, flicker is caused by varying load characteristics. Arc furnaces, motor starting, sawmills, and arc welding are typical sources of voltage fluctuations. Usually, these plants will be fed by a strong grid system and so flicker impact will be reduced. In the case of the wave station it is changes in the supply rather than the load which cause the voltage fluctuations. This coupled with the fact that an islanded system will normally have a weak grid system, makes it more susceptible to flicker problems. Figure 2 shows that for frequencies less than 1Hz, voltage fluctuations have little effect on flicker level. This is important in respect to a wave station whose voltage flicker frequency caused by individual wind waves would be of the order of.1hz. However, the modulation of the voltage envelope by these individual wave frequencies may just lie within the tolerable voltage limits for a weak grid. These large voltage excursions at low frequencies do have an effect on voltage flicker which must be assessed. Another effect of wind waves is that they tend to come in groups, producing a modulation effect on the individual waves. IEC STANDARD METER The IEC have set standards for the measurement and analysis of flicker severity. These are laid out in IEC 868 and show the design specifications of a flickermeter(4,5). The meter is essentially composed of several blocks, the purpose of each detailed as follows. Block 1 Input voltage transformer with automatic gain control this accepts a wide range of nominal mains voltages and adapts them to a maximum level compatible with the operation of the following circuitry. The output of the block is the RMS voltage modulated with the disturbance function. Block 2 Squaring Demodulator this circuit gives a component of its output a voltage linearly related to the amplitude of the fluctuation modulating the input. It is essentially simulating the flicker produced by a tungsten filament lightbulb. Part of the output component is the flicker waveform to be analysed, the other parts must be removed. Block 3 - Weighting Filters the first filter removes the dc component and the component at twice the mains frequency present at the output of the demodulator. The squaring function of the demodulator introduced this higher frequency. The output of this filter is solely the voltage flicker waveform. This filter comprises two filters back to back, a first order high-pass filter with a 3dB cut-off at.5hz and a low pass 6 th order Butterworth with 35Hz 3dB cut-off. The second filter weights the voltage fluctuation according to the lamp and human visual sensitivity response. The transfer function for the block is shown in the IEC standard. Block 4 Squaring multiplier and sliding mean function this performs two separate functions. Firstly the weighted flicker signal is squared to simulate the nonlinear eye-brain perception. Secondly the sliding mean averaging is performed to simulate the storage effect of the brain and the thermal capacity of the filament. This has the transfer function of a first-order low-pass filter with a time constant of 3ms. Block 5 Online statistical analysis this block samples the instantaneous flicker waveform at a suggested minimum of 5Hz. The flicker is classified into a minimum of 64 class levels and a cumulative probability function created. From this function it is possible to calculate a figure for the short term flicker severity, P ST, usually over 1 minute intervals. The long term flicker severity P LT is usually assessed over several days can be calculated from a successive of P ST s. The final design of the flickermeter is left to the engineer, although the standard does give tests to check that the meter falls within specification. SOFTWARE FLICKERMETER The purpose of the software flickermeter was to provide online and offline flicker performance for the LIMPET station.. The software flickermeter is divided into 2 parts, namely the instantaneous flickermeter model and the statistical analysis. Instantaneous Flickermeter Model The instantaneous flicker level is calculated by firstly standardising (reduction to per unit value) and then sampling the voltage waveform. This signal is squared, simulating the lamp response, before being passed to several filters which simulate the eye-brain response. This gives the instantaneous flicker sensation level. The output of the system is the instantaneous flicker level which must be further processed to give a value for P ST. Statistical Analysis The output of the instantaneous model produces a time trace of flicker sensation which is passed to a function that performs the statistical analysis. This must be done

4 in order to calculate the flicker sensation over a long enough period to allow for the build up of annoyance. The statistical method is because flicker is made up of regular and irregular voltage anomalies. These can make the instantaneous flicker level vary widely but perhaps not perceptibly over the observed period. This means that types of occurrence must be weighted. Also, it is important to know for what percentage of the observed period any given flicker level has been exceeded. In order to meet these requirements, a function was written which sorted the instantaneous flicker into a series of class levels, and then built up a probability function of class level exceedence which could be used to calculate the short term flicker severity. The shortterm flicker severity may be evaluated in as short a time as 1 minutes as a 1 minute interval is long enough for an individual to notice the disturbance and its persistence without giving too much importance to isolated voltage changes. Classification Method For the software flickermeter, the classification method chosen was to have a 2 level logarithmic classifier. The percentage of time the flicker spends in each of these levels is then calculated over the 1 minute test period. A cumulative probability function (CPF) is established from the class level data. The probability function is defined as the probability that a level will be exceeded. The P ST value is defined as P ST = K1P1 + K2P2+...K np n where P1, P2,.. Pn are CPF percentile points with an assigned probability of being exceeded and K1, K2,.. Kn are weighting coefficients. In order to find the flicker level at a particular percentile point it is necessary to use an interpolation algorithm. Details of possible methods are given in the IEC standard. The maximum flicker level was chosen to get a good spread of data throughout the levels, which in turn produces a smoother CPF for P ST calculation. The P ST could then be calculated using the equation with the percentiles and weightings given in Table 1. Percentile (P n ) Weighting (K n ) Table 1 Percentiles and Weighting Coefficients Percentile Interpolation There are several problems associated with the interpolation of the percentile points from the cumulative probability function. Interpolation is usually necessary as it is unlikely that all the required percentile points lie exactly on the CPF at measured values. The methods employed to deal with the problems experienced are shown below: Logarithmic classes The flicker level is sorted into classes or bands of flicker severity, these bands being spaced logarithmically. If a CPF was to be plotted solely from this information it would form a staircase function. This type of function cannot be used to accurately interpolate the percentile points. Instead, the midpoint of each class was found linearly and a curve fitted between the points. This produced a smooth CPF for more accurate interpolation. Empty Classes There is the possibility that there will be class levels which have no flicker points in them. When the CPF is constructed, even with the curve fitting described above, these empty classes produce flat spots, or part of a staircase function. If the required percentile happens to lie at this flat spot, the interpolation can have infinite solutions. To remove this possibility, any flat spots in the CPF are smoothed by curve fitting using data either side of the flat spot. The.1 percentile point This is the point in the function exceeded only by.1% of the flicker or 1 in 1 samples. If the maximum flicker occurs only occasionally then this point will not be measured. Statistically it must exist, but over the given sample time it may not have been recorded. To ensure that this percentile point can be found from the CPF, several points are used to extrapolate back to the.1 percentile point. The weighting applied to this percentile coefficient in the calculation of P ST is small compared with the other coefficients and so the accuracy of this method is adequate. Flickermeter Sample Frequency The IEC standard states that the flickermeter classifier must have a sample rate of at least 5Hz. However, due to the low frequency of sea waves and the inertia of the turbine/generator system, the frequencies produced by a wave-power station are much lower. The drawback of the standard flickermeter for real time analysis, is the high sample rate and the amount of processing required to produce a value for P ST. To store the voltage or power data for off-line analysis by the flickermeter model consumes a lot of memory. The solution is to reduce the sample frequency of the flickermeter, so that less storage and processing is required. Although the flicker created by the wavepower station is of a low frequency, reducing the sample rate of the classifier will cause problems with the percentile interpolation. Over a 1 minute interval at a frequency of 2Hz there will be only 2 samples and this greatly reduces the number of samples per class interval. This has a detrimental effect on how well the CPF can be produced and hence how accurately the percentiles can be interpolated for P ST calculation. In order to understand the effect that the classifier sample frequency had on the accuracy of the calculation, a test was devised to compare various

5 classifier set-ups at low frequencies. Several traces of voltage were processed with the flickermeter sample frequency ranging from 5Hz down to 1Hz. The results are shown in Figure 3. flicker analysis can be performed on individual phases and this could highlight conditions of grid unbalance. 23 PST % Error Flickermeter Sample Frequency R-ph Voltage (Volts) Y-ph Voltage (Volts) Figure 3 Classifier Sample Frequency Comparison This shows the worst case error experienced by the classifier over a range of classifier sample frequencies. Although the errors at 1Hz and 2Hz are small, it is thought that running at sample frequencies lower than 5Hz would introduce unacceptable errors. This is ideal in that this matches the data acquisition sample frequency on the LIMPET station. Testing and Calibration With the flickermeter and statistical classifier set up as described and with the sample frequency set to 5Hz, it was then tested using frequency and modulation amplitude values specified in the IEC standard. The flickermeter was tested with a range of frequencies from.5hz to 25Hz with modulation amplitudes set to give a P ST of 1. The results of the calibration test showed that for the standard input frequencies, the software flickermeter was performing to within 1% of the specification. DATA ANALYSIS The data acquisition system on the LIMPET plant has been set up to record various signals that describe station operation. These are to monitor, control and evaluate all aspects of the station s performance. The traces of interest for flicker measurement are the line voltages returned from the inverter system. At the time of publication, the LIMPET station has been operational for a short period only, and hence the results presented are only preliminary and have been taken whilst the station has been operating below its installed capacity of.5mw. The station in this commissioning condition is not running system controls which have been designed to improve the quality of supply (6). Figure 4 shows an example of the phase voltages measured whilst the station was producing an average power of 6kW, with peaks reaching 12kW. The data acquisition system for analysis purposes saves data in sets of between 1 and 15 minute periods which are ideal for flicker analysis. It must also be noted that B-ph Voltage (Volts) Time (secs) Figure 4 - Voltage Traces from LIMPET In order to find the flicker causing frequencies produced by a wave-power station, a spectral analysis of the voltage signals in Figure 4 was performed. These are shown in Figure 5. The spectra show that there is a spread between periods of 1 2 seconds, which are the naturally occurring wave periods at the LIMPET site. However, as the turbine/ generator combination has very high inertia, the effect of these individual waves is reduced. The spectra also show a frequency content of the order of 6-9 seconds. This is a modulating wave period that occurs due to a build up and then decline of successive waves occurring in groups. The inertia will have little effect on these longer periods as it only deals with short term storage. Figure 5 Spectra of Voltage Signals Flicker analysis The voltage spectra of Figure 5 show that there are a range of flicker causing frequencies at various intensities. Each of the phase voltages requires flicker analysis for each of the data sets under investigation. The instantaneous flicker produced by the flickermeter for a single voltage trace is shown in Figure 6.

6 Instantaneous Flicker Level 9 x Time (secs) Figure 6 Instantaneous Flicker with class levels This shows the variation in flicker intensity and the class levels used in order to classify it. The nature of the flicker variations show that the trace is more often at lower levels than at higher levels therefore being better suited to a logarithmic classifier. The resultant cumulative probability function produced from the classified instantaneous flicker sensation trace in Figure 6 is shown in Figure 7. This shows a smooth and complete function, a result of the having adequate data along with a logarithmic classifier. This type of CPF produces confidence in the interpolated percentile points used in the final calculation of P ST. Cumulative Probability Function % CPF of signal permanence in classes Class Number Figure 7 CPF with percentiles RESULTS A range of 3 phase voltage traces were available from the LIMPET station taken from the station operating at average power levels just under 1kW with peak power output reaching 15kW. The files were taken on successive days but all displayed similar frequency content of that shown in Figure 5. Also, since the station was in a commissioning mode when the data was being taken, all the power levels are very similar. This allows the files to be used as a group and average flicker levels to be found. The results are given in Table 2. The results show that for the range of data analysed, the flicker is well within acceptable limits, having an average of 14% of maximum tolerance. It can also be seen that the standard deviation is small, meaning that for the files analysed the flicker performance is similar. Phase Flicker P ST R-Phase.125 Y-Phase.152 B-Phase.144 Mean flicker Standard Deviation Table 2 Flicker Results A further result of the analysis is that the flicker values produced for each phase are slightly different. The values shown in Table 2 give the phase average over all the data analysed. The difference is to be expected since an islanded system such as Islay is unlikely to have a balanced grid and hence the phase susceptibility to flicker will be different for each phase. CONCLUSIONS It has been shown that it is possible to create and test a software flickermeter that complies to the IEC standards. Furthermore it has been shown that this meter can be used to assess impact on the grid that a wave-power station has. Frequency analysis of the voltage waveforms measured at LIMPET have shown that the frequencies contributing to flicker are those of the individual waves and a low frequency modulated wave. The frequency analysis also showed that because the voltage waveform was subject to only low frequency fluctuations, the flickermeter sample frequency could be reduced without loss of accuracy. The work to date has shown that the LIMPET station is working well within acceptable limits of flicker. Although the data was taken when the station was working below its installed capacity, it is thought that its flicker performance at rated output will be similar since a control strategy based on quality of supply will be used once the station has been fully commissioned REFERENCES 1. Heath T.V., The History and Status of the LIMPET Project, IMechE Seminar, November Description of Limpet Wave Station and Technical Specifications available on WaveGen s website: Beattie W.C., Sprevak D., Alcorn R.G., Producing acceptable electrical supply quality from a wave-power station, Proc Conference OMAE 98 Greece, IEC 868, Flickermeter Function and Design specifications, IEC, Geneva 1986 (BS EN 6868-) 6. IEC 868-, Flickermeter Evaluation of Flicker Severity, IEC, Geneva 1991 (BS EN 6868-) 7. Alcorn R.G., Beattie W.C., Control strategy development for remote Wave-Power Stations, UPEC 2 Conference, Belfast.