Corona noise on the 400 kv overhead power line - measurements and computer modeling

Transcription

1 Corona noise on the 400 kv overhead power line - measurements and computer modeling A. MUJČIĆ, N.SULJANOVIĆ, M. ZAJC, J.F. TASIČ University of Ljubljana, Faculty of Electrical Engineering, Digital Signal Processing Laboratory Tržaška 5, 1000 Ljubljana, Republic of Slovenia aljo.mujcic@ldos.fe.uni-lj.si Phone: Fax: Abstract The paper presents an experimental research in the corona noise performed on a 400 kv overhead power line with a horizontal disposition of conductors. The high-voltage (HV) power line in its role of a communication channel is a source of different noises characterizing an important feature of this communication media. The main objective was measuring and modeling the corona noise. Measurements of the Power Spectrum Density (PSD) and relative corona noise within a power frequency period are presented for different weather conditions. The algorithm for measuring variations in the corona noise level is described on the basis of noise samples. The maximum peaks exceed the Root-Mean Square (RMS) noise level by 7.95 db at foul weather conditions and by.9 db at fair weather conditions. An appropriate computer model of the corona noise compliant with the measurement results is proposed. Keyword: noise measurements, digital power line communication, corona noise, highvoltage power line. I. Introduction The corona noise is caused by partial discharges on insulators and in air surrounding electrical conductors of overhead power lines. Discharges occur on the three different phase conductors at different times. The corona noise level is considerably dependent on weather conditions. It is well known that the effect of the corona noise is particularly strong in foul weather conditions [1-5]. Characterization of the power line noise in international standards is based on the PSD and average RMS voltage [1, ]. The corona noise is defined as a voltage or power generated by dischargers on the power line [5]. Properties of the corona noise are significant for both analogue and digital power line communications but their influences on decreasing the channel quality are different. The analogue Power Line Carrier (PLC) modem is tolerant to degradation of the channel quality caused by the corona noise. Thus, the modeling of this noise was based on the Additive White 1

2 Gaussian Noise (AWGN) [1,]. On the other hand, the digital PLC modem depends on the Signal-to-Noise Ratio (SNR). If SNR is under the minimal required threshold, all services will drop out simultaneously []. The lowest SNR in digital PLC communications appears when the corona noise has one of the maximum values. Therefore, modeling of the corona noise is very important as a starting point for investigations into channel coding techniques in the digital PLC communications. In the analogue PLC communications the corona noise is considered as AWGN [1]. Its level is described with an average RMS voltage assuming that the power of the corona noise is equivalent to the white noise. In digital PLC communications such approach leads to wrong results []. From the nature of the corona noise it is clear that the noise has different value levels within a power frequency period. Typically, the measurement of the corona noise gives only its average level within an observed time interval. Our aim was to measure noise levels during a power frequency period and to model the noise level variations. The PSD and the ratio of the average RMS voltage at any moment in the period of 0 ms to the average RMS voltage of the corona noise give a complete description of the noise. The applied theoretical approach to this measurement is presented in []. Our adjustment of this algorithm for the corona noise measurement was made on the basis of digitally recorded samples (Section VI). The analysis of the relative corona noise within a power frequency period was made with recorded equidistant and peak noise samples. The paper is organized in the following manner. The basic corona noise is described in Section II. The measurement setup is given in Section III. The measurement of the noise PSD is presented in Section IV. Section V proposes an approach to the measurement and calculation of the corona noise variations inside a power frequency period. This section also addresses corona noise variations inside a power frequency period at foul and fair weather conditions. Section VI gives an algorithm for a mathematical approximation of the relative corona and background noise. A model of the corona noise is also presented in this section. Our conclusions are given in the final part, Section VII. II. Corona noise Ionization of air surrounding conductors of the HV power lines caused by electrostatic fields in these lines generates corona currents in the form of impulse pulses. Corona discharges, randomly distributed along the HV power lines, inject current impulses into bundle conductors [3,5]. Discharges on the three different phase conductors occur at different times.

3 When even the voltage on a particular phase is high enough, a corona burst occurs and a noise is generated. This discharge and the resulting noise occur primarily on the positive power line voltage wave. Therefore, the investigation of the corona noise in digital PLC communications has to deal with positive corona because of the small amount of the negative corona [3]. Although conductors are designed to minimize corona discharges, surface irregularities caused by damage, insects, raindrops or contamination may locally enhance the electric field strength sufficiently high for corona discharges to occur [3,5]. The corona noise level generated by overhead power lines suffers from various parameters such as: Atmospheric conditions, Line length, Average value of altitude, Size of conductors and their configuration, Type of connection, Bundle conductor composition, Voltage gradient, and Ground resistance. The corona generated currents change with atmospheric and also environmental conditions that cannot be defined accurately and are uncertain in nature. To deal with these generated currents, it is appropriate to represent it with a probabilistic model, which takes into account uncertainties of the above-mentioned parameters. Apart from atmospheric conditions, which have a predominant influence on the corona noise level and vary during the time, there are also some other factors that also affect the level of the noise PSD but they are almost constant in any weather condition and normal operation of the power line. Because of the nature of the corona noise, the HV power line as a communication channel does not represent an AWGN environment. III. Measurement setup and methodology III.I. Measurement setup A basic setup for the noise measurement is shown in Fig. 1. The PLC Line Trap Units (LTUs) are connected in series with the power line to contribute a correct transmission and reception of the power line carrier signals. The LTU impedance for the power line frequency is high and for the industrial frequency it is very low. A coupling circuit is used to connect the 3

4 measurement equipment to the HV power line and to protect the equipment from being damaged because of the 400 kv/50 Hz signal. Functions of the coupling device for noise measurement shift the noise from the power line into the measurement equipment, match impedances and electrical insulation between the HV power line and the measurement equipment. Fig. 1. Setup for the corona noise measurements The corona noise measurements were conducted at a substation on 3-phase, 50 Hz, 400 kv power line with a horizontal disposition of conductors. This type of the power line is common and can be found all over Europe. Noise measurements on this HV power line were made with an existing coupling circuit infrastructure for two-phase connection. Corona noise measurements were conducted with band pass filters shown in Fig 1. Amplitude characteristics of the four selected filters with a 3 db bandwidth are depicted in Fig. The selection of central frequencies and bandwidths are discussed in Section IV. 4

5 Fig.. Amplitude characteristics of the applied bandpass filter with: a) central frequency 199 khz and bandwidth khz b) central frequency 74 khz and bandwidth 0 khz c) central frequency 73 khz and bandwidth 34 khz d) central frequency 80 khz and bandwidth 90 khz III.II. Measurement methodology Within our investigation into the corona noise we measured: PSD, Average RMS voltage at the filter output of a given bandwidth, and Variations in the RMS voltage within a power frequency period. The spectrum/network analyzer HP 3589A was used for estimation of the noise PSD and measurements of interference from external sources. The measurements were conducted at fair and foul weather conditions using the spectrum/network analyzer at the input of a bandpass filter (Fig. 1) with 100 averages and a bandwidth of 1 Hz. The noise power level for different bandwidths was calculated by using the following expression [1] P Nf 1 = P Nf f + 10 log f 1 (1) 5

6 where: P Nf1 - noise power level with bandwidth f1 P Nf - noise power level with bandwidth f. The average RMS voltage was measured with the Fluke 199C Scope meter and verified with calculated values obtained with recorded samples (Section IV). To measure variations in the RMS corona noise within a power frequency period, waveform need to be recorded. This was done with the Fluke 199C Scope meter. The meter was controlled by a personal computer that also collects and stores all the measured data. The stored waveform points are samples of the measured signal and they are defined by the time and amplitude value consisting of a sequence of measured points. We have measured a single waveform consisting of: equidistant waveform samples and sequence of minimum and maximum waveform points of noise samples. VI. PSD of the HV power line noise Similarly as signals, the noise too can be described in a spectral domain. Because of its random nature, it is usually defined in terms of PSD that can be used as a measure of a continuous broadband noise and discrete peaks. PSD measures distribution of the noise power with frequency. Therefore, the noise PSD represents an average noise level in the frequency domain. In practice, we can only measure PSD in a finite band. Fig. 3 shows the noise level measured at high frequency cable in a substation. Results of the noise PSD measurements made at fair weather conditions are given in Fig. 3a. Using the same approach, the noise PSD was measured also at foul weather conditions. The obtained results are given in Fig. 3b. From Fig. 3 we can see that the following is characteristic for the measured PLC channel: corona noise; interference with other power line carriers (the highest peaks are in Fig. 3a). Foul weather conditions significantly affect the corona noise. The noise level at such conditions is approximately 15 db above the level at fair weather conditions (Fig. 3b). Also observed from Fig. 3 is that the corona noise PSD decreases with the increasing frequency. With the obtained measurement results average levels of the background noise and spectral components caused by other noise sources existing on the observed power line can now be determined. 6

7 Fig. 3. PSD of the noise measured at a high-frequency cable at fair weather (a) and foul weather conditions (b) To know how the frequency content changes over the period of the industrial frequency, which is not possible from PSD investigations into the corona noise should be envisaged in the frequency band with no spectral components from other sources of interference such as for example PLC systems, radio navigation systems and broadcast radio stations. Therefore, to determine the corona noise behavior during a power frequency period, appropriate filters are needed. The selection of the central frequency and bandwidth of such filters depends on PSD of an actual power line and a transfer function of both the coupling circuits and the LTU (Fig. 1a). The filtered signal is composed of a continuous background and corona noise. V. Dependence of the corona noise on the voltage of power frequency The average RMS voltage U NRMS is computed by squaring the instantaneous voltage of noise, integrating over the desired period, and taking the square root: U NRMS = 1 T T ( un ( t) ) dt [ V ] 0 () 7

8 We compared the RMS measurements made with the Fluke 199C Scope meter and the values calculated from the measured waveform. The average RMS of this estimation at fair and foul weather conditions is presented in Table 1. The RMS computation of the corona noise gives the average power carrying capability. The RMS measurement of a random noise value requires a large number of samples and a long averaging time to allow for an accurate estimate. From the nature of the corona noise described above, we can conclude that noise level within a power frequency period isn t constant. Our objective is to describe these changes. Variations in RMS of the corona noise voltage within a power frequency period can t be obtained using equation (). Such stochastic process can be described as a white noise with a variable variance or RMS within a power frequency period []. Table 1 No. Central frequency Fair weather conditions Foul weather conditions of the filter [khz] Bandwidth [khz] V [mv] V [dbv] V [mv] V [dbv] We analyzed a single waveform consisting of equidistant waveform samples and a sequence of minimum and maximum waveform points of noise samples. The maximum or minimum noise samples (positive and negative peaks) exceed the noise power average value. An average excess increases with the length of time over which the peak is observed. The aim is to obtain the ratio of the average RMS voltage at any moment within a 0 ms period to the average RMS voltage of the corona noise by applying the classical approach using equation (). The average RMS voltage at any given moment in a 0 ms period is obtained by averaging samples from successive periods of the industrial frequency at a corresponding sample location (Fig. 4) (e.g. sample 1, M+1, M+1 N-M+1). Our analysis of the relative corona noise within a power frequency period was made with Matlab on captured data. The stored data comprise a sequence of N samples for each filter and information of applied sample rate Fs. The N samples may be subdivided into N = int M L (3) 8

9 blocks composed of M 3 = TF S = 10 F S 0 (4) samples per power frequency period T. Thus, the blocks are represented as u Nk ( n ) = u N ( k + nm ) n = 0, 1,...,M 1 k = 0, 1,...,L 1, (5) A graphical illustration of these segments is shown in Fig. 4. Fig. 4. Relative corona noise algorithm The relative corona noise as a ratio of the average RMS voltage at any given moment in a 0 ms period to the average RMS voltage is defined as L 1 i= 0 ( i) k U = L NRMS N 1 k = 0,1,... M 1. (6) u ( i) i= 0 u Nk N N where N is number of samples, L number of power frequency periods and M is number of samples per power frequency period T. Equation (6) gives variations in the RMS noise voltage within a power frequency period. Fig. 5 shows relative corona noise variations within a power frequency period for different bandwidths of the receiving filters and foul weather conditions. The amplitude characteristics of the used filters are presented in Fig.. As seen from Fig. 5, the relative corona noise has three peaks and represents the corona noise contributions from three conductors [,4]. The 9

10 relative corona noise is normalized where one unit corresponds to the average RMS noise voltage. Fig. 5. Relative corona noise at foul weather conditions for filters of Fig. a) central frequency 199 khz and bandwidth khz b) central frequency 74 khz and bandwidth 0 khz c) central frequency 73 khz and bandwidth 34 khz d) central frequency 80 khz and bandwidth 90 khz There are two important observations concerning the obtained results. Firstly, the relative corona noise changes with the voltage of the power frequency during a 0 ms period whereas the relative corona noise retains the same shape. Secondly, the ratio between the three peak values of the corona noise envelope doesn t vary with the central frequency and the selected filter bandwidth. The same results are obtained from the ratio of the average RMS voltage of peak measurements at any moment in a 0 ms period to the average RMS of the peak measurements obtained by a classical approach. Fig. 6 shows variations in the corona relative noise based on peak measurements at foul weather conditions. If we use a peak detector to measure the noise, then the average RMS noise level will be lower than the RMS peak level. 10

11 The relative corona noise is the ratio between the RMS voltages at any moment to the average RMS voltage that is calculated using all received samples. Thus, the difference between the average power of peak measurements and the average power of sampled measurements is eliminated in the relative corona noise. RMS of the corona noise depends on the power frequency voltage resulting in a burst of short trains of impulses with a fundamental burst repetition frequency 150 Hz [1-4]. From our measurements we established that the maximum peaks exceed RMS noise level by 0 log(. 5 ) = db. (7) From these results we can conclude that an instantaneous power peak during the occurrence of corona impulses can cause bit or burst errors in data transmission. Fig. 6. Relative corona noise at foul weather conditions using peak detection a) central frequency 199 khz and bandwidth khz b) central frequency 74 khz and bandwidth 0 khz c) central frequency 73 khz and bandwidth 34 khz d) central frequency 80 khz and bandwidth 90 khz 11

12 Relative corona noise variation within a power frequency period at fair weather conditions is presented in Fig. 7. We can see three peaks that represent the ratio of the maximum voltage peak to the average RMS of the corona noise. Our measurement results show that during fair weather conditions the effect of the corona noise is small and maximum peaks exceed the average RMS noise level by 0 log( 1. 4 ) =. 9 db. (8) The average noise level at fair weather conditions is smaller than the average noise level at foul weather conditions. The maximum level of the corona noise at fair weather is smaller than the minimum level of the corona noise at foul weather. Therefore, fair weather results are not crucial when a design of a digital PLC modem is considered. Fig. 7. Relative corona noise at foul weather conditions using peak detection at fair weather conditions: a) central frequency 199 khz and bandwidth khz b) central frequency 74 khz and bandwidth 0 khz c) central frequency 73 khz and bandwidth 34 khz d) central frequency 80 khz and bandwidth 90 khz 1

13 VI. Modeling of the corona noise Before envisioning hardware implementation, suitability of all algorithms in communication systems must be proven by computer simulations. The overhead HV power line as a communication channel can be modeled as a time varying frequency-dependent channel [,6-8]. Modeling of a power line channel requires an appropriate model of the corona noise. The first step is modeling a relative corona noise by a mathematical function. The dependence of an instantaneous corona noise voltage to 50 Hz power frequency can be approximated by three cosine signals with a period T1=1/(3*50) seconds and amplitude [] Voltage Ui max π Ui ( t ) 1 cos t 0 < t < T1 T 1 =. (9) U i max Ui max represents a maximum RMS value on a particular phase of a power line. Fig. 8a shows an approximated relative corona noise voltage in the time domain for the center phase to the outer phase coupling of a 400 kv line. From the known dependence of an instantaneous corona noise voltage on the 50 Hz power frequency the average RMS value is defined as [] T 1 1 U Nrms Ui ( t ) dt (10) 3T = 3 i= Note the difference between the measurement results and the approximated dependence of the relative corona noise. The equation (9) doesn t include the continuous background noise. To allow for comparability with the measurement results and appropriate modeling of the background noise, U min is added in equation (9) Ui max U min π Ui ( t ) = 1 cos t + U min 0 < t < T1 T. (11) 1 Fig. 8b shows the approximated relative corona noise voltage over time for the center phase to the outer phase coupling of a 400 kv line with included background noise can be determined from the relative corona noise obtained by measurements. U min. This level 13

14 Fig. 8. Model of the relative RMS noise voltage dependence on time within a power frequency period: with no background noise (a) and with background noise b) The background and corona noise can be synthesized by filtering the white noise source and multiplying it by function U ( t ) which describes its dependence on the power frequency. The influence of weather conditions is represented by a separate block average RMS and ratio of three peaks. The average RMS and the ratio of the three peaks block have two outputs. The first output is the average RMS of the corona noise which multiplies samples of the white noise. The second output holds values of relative corona noise amplitudes particular phase for the block dependence on the power frequency. The noise-shaping filter can be described by H( z ) 1 = n 1+ a i z i= 1 i. (1) U i max on a The parameters of the shaping filter can be determined from a measured noise signal using a parametric AR estimator. Fig. 9 Generation of the background and corona noise 14

15 VII. Conclusions In this paper we present results of our investigation into the relative corona noise dependence on the power frequency voltage in HV power line communications. Our approach was based on results of our measurements, as well as those of statistical analyses of the captured data and modeling at different weather conditions. By measuring PSD we determined the various types and levels of noise sources existing on the power line. We established that the relative corona noise describes the corona noise behavior during a power frequency period. From results of our measurements and those of statistical analyses we can conclude that the corona noise cannot be described as a white noise with a constant variance. The results of our measurements which were conducted at foul and fair weather conditions, show that the character of the corona noise is similar to the one of the Gaussian noise with a variable variance during a power frequency period. The statistical analyses prove that the signal-tonoise ratio varies during a power frequency period at all weather conditions but is higher at foul weather conditions than at fair weather conditions. As a consequence of the SNR variations the probability of error is also variable and considerably affects the reliability of digital communications. Burst errors in digital PLC communications are possible when the envelope of the corona noise occupies one of the maximum values. Based on thus attained results, a computer model of the corona noise is proposed (Fig 9). These attained results are very important as they provide solid basis for investigation into channel coding techniques in the digital PLC communications. Acknowledgement The authors of the paper wish to acknowledge the support of the Ministry of the Economy of the Republic of Slovenia and Iskra Sistemi while performing this work under the project Digital Power Line Carrier Communications. References 1. INTERNATIONAL ELECTROTECHNICAL COMMISSION, IEC REPORT (1980) Planning of (single-sideband) power line carrier systems. Geneva. CIGRE STUDY COMMITTEE 35 Working Group 09 (000) Report on Digital Power Line Carrier. 15

16 3. P. Sarma Maruvada (000) Corona performance of high-voltage transmission lines. Research Studies Press Ltd. Philadelphia 4. An American National Standard (1980) IEEE Guide for Power-Line Carrier Applications. The Institute of Electrical and Electronics Engineers Inc. New York 5. S Cristina, M. D Amore (1985) Digital Analytical Method for Calculating Corona Noise on HVC Power Line Carrier Communication Channels. IEEE Trans. on Power Apparatus and Systems, vol. PAS-104, No. 5, pp N.Suljanović, A Mujčić, M Zajc and J F Tasič (003) Power line tap modeling at power-line carrier frequencies with radial-basis function network. Engineering Intelligent Systems vol.11, No. 1, pp MUJČIĆ, Aljo, SULJANOVIĆ, Nermin, ZAJC, Matej, TASIČ, Jurij (00) Detection of nonlinearities in communication channel phase characteristics. Proceedings of the eleventh International Electrotechnical and Computer Science Conference ERK 00, ISSN ,pp , Portorož, Slovenia. 8. SULJANOVIĆ, Nermin, MUJČIĆ, Aljo, ZAJC, Matej, TASIČ, Jurij (00) Tapped power-line modelling with radial function network. Proceedings of the eleventh International Electrotechnical and Computer Science Conference ERK 00, ISSN , pp , Portorož, Slovenia. 16