High Impedance Fault Detector

This invention relates to a method and means for detecting a high impedance fault or arcing phenomenon in an alternating current electrical distribution system.

For the purpose of supplying uninterrupted power to consumers, protection systems must isolate any type of electrical fault which may occur in the electrical distribution system under various conditions. In modern distribution systems, protective relay means have been developed to cope with most circumstances, and these systems are able to respond rapidly to sudden and sometimes violent changes in current and voltage caused by short circuits. However, the problem of high impedance faults presents a major concern since many relay systems are unable to respond quickly and accurately under such conditions. A high impedance fault is a type of electrical fault which commonly occurs in a distribution feeder when the fault current flows through an equivalent high impedance which is caused by an electric arc. Because high impedance faults are often associated with electric arcs they are sometimes called an "arcing fault".

A typical single phase high impedance fault occurs when a plant such as a tree branch contacts a high voltage distribution power line, which often results in an arc between the plant and the power line. From the point of view of a protective relay situated at the distribution point for the power line the arc between the plant and the power line will present a high impedance. Further, if the protective relay is not able to detect the fault the result may be quite catastrophic, for example resulting in a bush fire.

Protective relay systems are often not able to detect such high impedance faults because fault currents are often far too small to be recognized by conventional relay systems. There are often no rapid nor ample changes in voltage or current waveforms when high impedance faults occur because the impedance from the power line to ground is quite high.

It is common for low current, high impedance faults to occur on distribution feeders and other relatively low voltage power lines. A high impedance fault usually occurs when a distribution line conductor is contacted by a high impedance grounded object, such as a tree, or when a conductor breaks and falls onto a poorly conducting earth surface. A great deal of research in the past has concentrated on creating more sensitive ground fault detection devices, since many researchers believed that high impedance faults were often not detected because the protective relays were not sensitive enough. Consequently, much research has gone into the design of very sensitive over current relays, in particular zero sequence over current relays for distribution networks. Although this research has improved grounded fault protection systems, the problem of low current faults remaining undetected has remained, the reason for this being that these devices may not distinguish fault currents below normal load levels.

Recent research has investigated the characteristics of high impedance faults and the use of signals other than fundamental frequency signals to detect high impedance faults. It was found that one of the most important characteristics observed during arcing faults is that harmonic signals increase to a measurable value which may be detected during occurrence of a fault. For example, research at Texas A & M University has addressed the characterisation of fault currents associated with downed conductors. From this fundamental understanding of low current fault characteristics a technique was developed which utilised high frequency (two kilohertz to ten kilohertz) current components for fault detection. This technique is based on the observation that arcing was almost always associated with downed conductors resulting in a modulated current waveform rich in high frequency components. The high impedance fault detector therefore utilised high frequency components to detect the occurrence of a high impedance fault, and operated in a satisfactory manner under certain operational conditions. However, field tests have shown that these high frequency signals can often not propagate past capacitor banks which are present in electrical distribution systems, and hence the system is unable to perform in many practical applications.

Further research on high impedance faults has shown that harmonic content during a high impedance fault is dependent upon many parameters. The harmonic content during a fault can be effected by many physical and environmental variables such as feeder configuration, shape of fault point, the manner in which a fault performs, load type and level and the prevailing weather. Using a single detection parameter such a high frequency signal, a low frequency signal, or a third harmonic frequency signal has been shown to have difficulties in detecting high impedance faults under certain conditions. Therefore, the most recent research has concentrated on detecting high impedance faults using multiple detection parameters, however this research has also indicated that arcing faults are complicated phenomena, the behaviour of which is influenced by a large number of physical variables and environmental parameters which are difficult to quantify. In this light, leading researchers C. J. Kim and V. Don Russell from Texas A & M University, in a report entitled "A learning method for use in intelligent computer relays for high impedance faults" (IEEE Transactions on Power Delivery, Volume 6, No. 1, January 1991 pages 109 to 115) stated that "the behaviour of the parameters of high impedance faults is very random and, in most cases, unpredictable".

The present invention, in summary, provides a method for detecting a high impedance fault or arcing phenomenon in an AC electrical distribution system having a fundamental frequency, comprising:

a. monitoring electrical signals in said electrical distribution system;

b. isolating a plurality of predetermined frequency components from the monitored electrical signals;

c. scaling the magnitudes of said predetermined frequency components according to selected scaling factors;

d. comparing said sealed magnitudes from a first data acquisition period with the scaled magnitudes from a preceding data acquisition period; and

e. detecting the occurrence of a high impedance fault or arcing phenomenon on the basis of said comparison.

The step of comparing said scaled magnitudes may comprise comparing the sum of said scaled magnitudes during said first data acquisition period with the sum of the scaled magnitudes during said preceding data acquisition period.

Further, the step of scaling the magnitude of a given predetermined frequency component may comprise determining and summing the number of sub-periods of a data acquisition period during which the given frequency component is greater than a selected value, and multiplying the resulting sum by a scaling factor corresponding to said given frequency component.

Preferably the detection of the possibility of a high impedance fault is determined by a comparison of the scaled magnitudes during first, second and third sequential data acquisition periods.

Preferably one of said predetermined frequency components is twice said fundamental frequency.

Preferably said plurality of predetermined frequency components comprises eight voltage and/or current signals.

The plurality of predetermined frequency components may include zero sequence current and voltage signals, and these zero sequence current and voltage signals may be utilised to detect solid ground faults, particularly in the case that the electrical feeder line comprises a three phase, four wire line.

In order, for example, to prevent damage occurring as a result of the high impedance fault, the method may also include a step of issuing a trip signal when the possibility of a high impedance fault is detected, to cause said at least one electrical feeder line to be isolated from the source of said AC electricity.

In accordance with the invention there is also provided a method for detecting an arcing phenomenon in an electrical distribution system comprising the steps of:

monitoring n selected frequency components C.sub.n of the voltage and current at a point on the distribution system;

generating a first pattern M.sub.j (C.sub.n) representative of the magnitudes of the selected frequency components during a first data acquisition period;

comparing the first pattern with a previously stored pattern M.sub.j-1 (C.sub.n) to determine whether the difference therebetween exceeds a pattern threshold Mt (C.sub.n) such that

determining the occurrence of an arcing phenomenon in the system on the basis of the comparison.

The present invention also provides a method for detecting the possible existence of a high impedance fault in an electrical distribution system conveying AC electricity of a fundamental frequency, comprising:

monitoring electrical signals in said electrical distribution system;

isolating a plurality of predetermined frequency components from the monitored electrical signals;

sampling the frequency components during at least one data acquisition period;

forming and storing a first pattern corresponding to the samples during said at least one data acquisition period;

comparing said first pattern with a previously stored template pattern; and

determining the possible existence of a high impedance fault on the basis of said comparison.

The template pattern may comprise a calculated threshold pattern or may comprise a pattern formed and stored from samples of said frequency components during a sequentially preceding data acquisition period. Where the template pattern comprises a previously sampled and stored pattern, a step may be provided of forming a second pattern from sampling the frequency components during a further data acquisition period, and comparing said template pattern with both said first and second patterns to determine the possible existence of a high impedance fault.

A step of confirming the existence of a high impedance fault may also be included, comprising analysing a sequence of chronologically sampled data from at least one of said plurality of frequency components to determine whether the magnitudes of the at least one frequency component is modulated at twice said fundamental frequency.

The present invention also provides a high impedance fault or arcing phenomenon detector for use in an electrical distribution system including at least one electrical feeder line for conveying AC electricity of a fundamental frequency, said detector comprising:

a. monitoring means for monitoring electrical signals on said at least one electrical feeder line;

b. filtering means for isolating a plurality of predetermined frequency components from the monitored electrical signals;

c. weighting means for scaling the magnitudes of said plurality of predetermined frequency components according to selected scaling factors;

d. comparison means for comparing said scaled magnitudes from a first data acquisition period with the scaled magnitudes from a preceding data acquisition period; and

e. detecting means for detecting the occurrence of a high impedance fault or arcing phenomenon on said electrical feeder line on the basis of said comparison.

An electrical isolation switch is also provided by the present invention for use in an electrical distribution system including at least one electrical feeder line for conveying AC electricity of a fundamental frequency, comprising:

a. monitoring means for monitoring electrical signals on said at least one electrical feeder line;

b. filtering means for isolating a plurality of predetermined frequency components from the monitored electrical signals;

c. weighting means for scaling the magnitudes of said plurality of predetermined frequency components according to selected scaling factors;

d. comparison means for comparing said scaled magnitudes from a first data acquisition period with the scaled magnitudes from a preceding data acquisition period;

e. detecting means to detect the possibility of a high impedance fault on said electrical feeder line on the basis of said comparison and to generate a trip signal if the possibility of a high impedance fault is so detected; and

f. switching means to electrically isolate said at least one electrical feeder line from the source of said AC electricity upon generation of said trip signal.

The present invention also encompasses an electrical distribution system comprising:

a source of AC electricity of a fundamental frequency;

at least one electrical feeder line coupled to said source by way of switching means capable of electrically isolating said at least one electrical feeder line from said source upon receipt of a trip signal;

monitoring means for monitoring electrical signals on said at least one electrical feeder line;

filtering means for isolating a plurality of predetermined frequency components from the monitored electrical signals;

weighting means for scaling the magnitudes of said plurality of predetermined frequency components according to selected scaling factors; and

comparison means for comparing said scaled magnitudes from a first data acquisition period with the scaled magnitudes from a preceding data acquisition period, and to issue a said trip signal to said switching means if said comparison indicates the possibility of a high impedance fault on said at least one electrical feeder line.

In order to more easily facilitate the comparison of the scaled magnitudes during sequential data acquisition periods the scaled magnitudes of the plurality of predetermined frequency components during each data acquisition period are preferably summed, and the sums of said scaled magnitudes are compared by said comparison means.

The apparatus provided by the present invention preferably includes a sample and hold device, a threshold device, and a summing means, whereby the scaling of the magnitude of a given predetermined frequency component may be conveniently achieved by determining and summing the number of times during a data acquisition period in which the sampled magnitude of the given frequency component exceeds a selected threshold, and multiplying the resulting sum by a scaling factor corresponding to said given frequency component. The summing and scaling of the predetermined frequency components may be conveniently carried out according to the following formula: ##EQU1## Where: X is the weighted sum of the frequency components,

n is the number of frequency components,

W.sub.i is the scaling factor for the frequency components,

m is the total number of samples during a data acquisition period,

S.sub.ij is the sample magnitude for the frequency components, and

T.sub.i is the individual threshold for each frequency component.

To enable a better understanding of the working of the present invention a preferred embodiment thereof is described in detail hereinafter, by way of example only, with reference to the accompanying drawing, wherein:

FIG. 1 shows a diagrammatic representation of a typical high impedance arcing fault;

FIG. 2 illustrates a characteristic of high impedance arcing faults;

FIG. 3 illustrates sensing circuits of the preferred embodiment of the present invention;

FIG. 4 shows a block diagram of the apparatus of the preferred embodiment;

FIG. 5 shows a more detailed block diagram of the circuitry of the preferred embodiment;

FIG. 6 illustrates an example of a filtering circuit of the preferred embodiment;

FIG. 7 illustrates sampling and digital conversion circuits;

FIG. 8 illustrates a digital numeric processing arrangement for use in the preferred embodiment;

FIG. 9 is a block diagram of output circuits of the preferred embodiment;

FIGS. 10 and 11 illustrate the use of pattern recognition in the preferred embodiment;

FIG. 12 shows an example of sampled waveforms during an arcing fault;

FIG. 13 is a simplified flow chart illustrating the process utilised by the preferred embodiment;

FIG. 14 is a more detailed flow chart of the process flow chart illustrated in FIG. 13; and

FIGS. 15a to 15f illustrate an exemplarily diagrammatic pattern comparison process for the fault situation illustrated in FIG. 12.

FIG. 1 is a schematic diagram of a typical electrical distribution system 1 which may be affected by high impedance arcing faults. A source of AC electricity 2 is coupled through a transformer 4 and protection circuitry 3 to an electrical feeder line 6 which terminates at an electrical load 9. The feeder line 6 may typically be a three phase, three or four wire distribution feeder, or may comprise a single phase two or one wire line. As illustrated in FIG. 1, at some point along its length, the feeder line 6 is so close to a grounded high impedance object, such as a tree 10, that at least one wire of the feeder line 6 may to come into contact with the tree 10. This is indicated in FIG. 1 at a fault point 8, causing a high impedance fault between the feeder line 6 and the high impedance object 10, by virtue of an electrical arc therebetween. Because the arc currents in such faults are often quite small, such faults often escape tripping over current protection devices of the protection circuitry 3. However, even relatively small arcing currents can generate a large amount of localised heat build up, which may in turn result in an undesirable bushfire.

Detailed analysis of the electrical response of electrical distribution systems to arcs has revealed that signals of many different frequencies other than the fundamental frequency of the electrical system are generated during arcing. Tests have revealed that harmonic signals generally increase during the occurrence of a high impedance fault, and that the amount of increase of harmonic frequencies varies under different power system and fault conditions. For example, the harmonics generated by a high impedance arcing fault have been shown to be affected by physical and environmental variables such as feeder configuration, shapes of objects at the fault point, fault nature, load type and level, and prevailing weather.

The shape of objects at the fault point, such as tree or concrete road, is an important factor which affects harmonic content. If the fault point is sharp, it may result in point to point arcing, however if the shape is flat then arcing may result over a large surface area. These different types of arcing configurations have been shown to generate different levels of frequency signals.

Capacitor banks (not shown) which may be connected at one end of a feeder line can also affect the frequency signals detectable at the point of protection circuitry 3. High frequency signals, for example, can be shunted easily by capacitor banks, which present very low impedance to such high frequency signals. On the other hand, low frequency signals can pass capacitor banks much more easily.

It has also been found that the frequency components which appear as a result of an arc appear only intermittently, modulated at twice the fundamental AC frequency. The cause for the intermittent nature of frequency components resulting from an electrical arc can be conveniently described with reference to FIG. 2. The waveform 12 illustrated in FIG. 2 shows a sinusoidal source voltage waveform which may be found in an electrical distribution system under normal operating conditions. The curve 14 shows the voltage waveform which occurs as a result of an arcing fault. As shown in FIG. 2, arcing begins at points A1 and A2 near the voltage peaks of the sinusoidal waveform, and cease at points B1 and B2 respectively, when the voltage becomes too low to sustain the arc. This results in two arcing periods 16 during which arcing current flows to ground, causing the feeder voltage to noticeably drop as can be seen in FIG. 2 where the curve 14 is lower than the corresponding curve 12. Arcing occurs twice for every single electrical cycle, resulting in a characteristic harmonic frequency which is twice the fundamental frequency.

In consideration of the above, the preferred embodiment of the present invention presents a system for detecting the presence of an arcing fault by utilising the electrical characteristics of such high impedance faults, namely the presence of a range of frequency components other than the fundamental frequency, and the intermittent nature of the presence of such frequency components during a high impedance fault. A group of frequency components for use by the preferred embodiment has been found in which the combination of detected components can realise a high accuracy of fault detection, in spite of the fact that each component when considered alone has drawbacks for detecting faults. In particular, the preferred embodiment determines the possibility of a high impedance arcing fault on the basis of eight signals, namely:

V.sub.0 : zero-sequence (50 Hz) voltage,

V.sub.2 : second harmonic (100 Hz) voltage,

V.sub.s : eight harmonic (400 Hz) voltage,

V.sub.h : high frequency (2 KHz-10 KHz) voltage,

I.sub.0 : zero-sequence (50 Hz) current,

I.sub.1 : low frequency (120 Hz-130 Hz) current,

I.sub.3 : third harmonic (150 Hz) current, and

I.sub.s : eight harmonic (400 Hz) current,

In order to distinguish between signals present during normal system operation and those resulting from arcing faults, the selected signals are based on frequency components which have been found to show a significant increase when arcing faults occur under a wide range of different system conditions. Four voltage signals and four current signals are used as the detection criteria in the preferred embodiment.

Zero sequence voltage and zero sequence current signals have traditionally been used as parameters to detect unbalanced faults. It has been found that these signals are not sufficient alone for detecting arcing faults, but can add valuable information to assist the final detection of an arcing fault in a four wire feeder. In a well balanced power system, zero sequence voltage and current may increase when arcing faults occur. Additionally, zero sequence voltage and current may be used to detect the direction or position of a fault by comparing the angle between two phases.

Low frequency signals ranging from 120 Hz to 130 Hz have also been found to be of significant importance in detecting arcing faults. The relative increase in the magnitude of signal frequencies in this range is large enough to be easily detected under arcing fault conditions, since these signal frequencies should be zero under normal operating conditions. Another advantage is that signals at these low frequencies will exhibit little attenuation from capacitor banks present in the distribution system.

A third harmonic current component is useful and a detection criteria, particularly if the distribution loads are connected in delta configuration and for three wire distribution circuits. Although the third harmonic current can exist in relatively high amplitudes even under normal system operating conditions, it has been found that the increase of the amplitude during an arcing fault is sufficient to assist detection.

Eighth harmonic voltage and current signals have also been found to increase remarkably during arcing faults. Traditionally, odd harmonics (third, fifth and so on) have been considered to be the best indicators for arc detection, since they are predominant in amplitude during arcing. However, tests leading to the construction of the preferred embodiment have unexpectedly shown that the relative amplitude increase of even harmonics, especially the eighth harmonic can be larger than the relative amplitude increase of many odd harmonics. This results in the eighth harmonic being one of the best indicators of arcing faults under all system conditions.

High frequency current components includes all signals in the range of approximately 2 kHz to about 10 kHz which exists in the protected power system. It has long been known that high frequency current signals, particularly above 2 kHz increases significantly under arcing conditions. A disadvantage of such high frequency signals is that they cannot generally propagate long distances (greater than about 20 miles), particularly through feeders provided with filtering capacitor banks.

The eight selected frequency components each perform differently during an arcing fault, although in general all of them exist and increase during an arcing fault. Furthermore, some frequency components may increase more in a first power system condition, whilst others increase more in a second system condition. For example, in a distribution power line provided with filtering capacitor banks, the amplitude of the second harmonic signal may be detected to increase relatively higher than the high harmonic signals which get easily shunted to ground through the capacitor banks. The preferred embodiment, therefore, employs scaling or weight factors for each signal, which may be altered depending upon the system conditions prevalent at the particular distribution feeder.

For example, in a feeder provided with shunting capacitor banks the weight or scaling value for the second harmonic signal should be set higher than the weight value of the high frequency signals, since the second harmonic signal will increase relatively greater during an arcing fault. The ability to change weights for the individual signals allows the high impedance fault detector to be adjusted according to the distribution system upon which it is to be employed. The adjustable weighting values in the preferred embodiment also enables the fault detector to be easily readjusted after installation, to adjust for alterations in the distribution system, or to provide an opportunity to improve the accuracy of detection by learning from the results of previous arcing faults.

FIG. 3 illustrates a typical three wire three phase distribution feeder 18, and the preferred arrangements for monitoring the voltage and current frequency components utilised by the preferred fault detector. A voltage monitoring circuit 20 comprising a three phase capacitor voltage transformer 21 senses voltage signals V.sub.a, V.sub.b and V.sub.c, whilst a current monitoring circuit 22 comprising a three phase current transformer 22 senses current signals in the three phase wires.

FIG. 4 shows a block diagram of the preferred high impedance fault detector 34, which comprises five main components: input circuits 24, active filter system 26, digital processing circuitry 28, a micro processor control unit 30 and output circuits 32. The function of the fault detector 34 is to detect the presence of high impedance faults based on processing of received voltage and current input signals. In summary, signals are derived from the distribution feeder 18 by way of the voltage monitoring circuit 20 and current monitoring circuit 22, and are reduced to a level which is suitable for measurements. Frequency components are selected from the derived signals by individual filters of the active filter system 26, and are passed to the microprocessor control unit 30 after being sampled and digitally converted by the digital processing circuitry 28. Analysis of the input signals takes place in the microprocessor control unit 30, following a detection scheme which is explained in detail hereinafter. The results of the analysis is then utilised to operate a control system such as isolating relays by way of the output circuits 32.

The input circuits 24 of the detector 34 receive the voltage and current signals sensed by the voltage monitoring circuits 20, 22, and act to reduce the voltage and current signals to levels which can be comfortably handled by low voltage circuitry. The level of reduction may be accomplished by any convenient method, and in the preferred embodiment comprises step down transformers (not shown) and resistors (not shown) for the voltage signals, and for the current signals comprises a step down transformer (not shown) and an op-amp (not shown) coupled to convert current signals to more convenient voltage signals.

FIG. 5 shows a more detailed block diagram of the fault detector 34, illustrating the components of the active filter system 26. Each signal component, except the zero sequence components which are at the fundamental frequency, are selected by individual filters 36. FIG. 6 shows an example of a band pass filter 36 which is utilised in the preferred embodiment. The filter illustrated in FIG. 6 is a fourth-order band-pass Chebyshev cascaded biquad filter, which exhibits excellent tuning features, and is capable of obtaining Q factors of up to 100 or more. The individually filtered frequency components are then put through an amplitude adjustment stage 38, before passing to the digital processing stage 28. Although the analog filters utilised in the described embodiment perform well, digital filters may alternatively be used if required.

FIG. 7 illustrates a block diagram of the digital processor 28 selected for the preferred embodiment, which is based on products ADC0808 or ADC0809 of the National Semiconductor Corporation. The eight analog frequency component signals from the active filters are inputted to an eight-to-one analog multiplexer 50 which is clocked by signals from the microprocessor controller 30. The signal output of the multiplexer 50 is passed to a sample-and-hold circuit 52, which samples the analog impulse signal and passes the sampled magnitude to an analog-to-digital conversion stage 4. The sampled analog signal magnitude is there converted to a digital number, which is passed to the microprocessor controller 30 for further processing and analysis.

The microprocessor controller 30 is based upon a Motorola M6802 microprocessor 56 and a block diagram of the main components of the controller 30 is illustrated in FIG. 8. Briefly, the microprocessor 56 coupled to non-volatile memory 58 and read/write RAM memory 60, and also to a programmable input output port device 62. The digital processing and analysis procedures which are carried out on the digitised frequency component signal data by the microprocessor are stored as instructions in the non-volatile memory 58, whilst the RAM memory 60 is used for storing patterns indicative of the operational history of the distribution system. The programmable input output port 62 is used both to receive the digitised frequency component signals from the digital processor 28, and also to issue instructions to the output circuitry 32.

After the input data have been analysed by the microprocessor, instructions may be issued to the output circuitry 32. The function of the output circuitry 32 is to link the microprocessor control system 30 with the control system in the electrical distribution sub-station, such as isolating relays for isolating faults on the relevant distribution feeders. A block diagram of the output circuitry 32 is illustrated in FIG. 9, which comprises decoding circuits 64, relay driver circuitry 66, and message indication circuits 70. The instructions from the microprocessor controller 30 are translated by the decoding circuits, which may provide messages to the operators of the distribution system by way of the message indication circuits 70 which may include indicator lights indicating normal or abnormal operation of the feeder line. When analysis of the input data indicates a high impedance arcing fault, the fault detector 34 is also able to trip the circuitry breaker or relay of the feeder by way of the relay driver circuits 66, in order to isolate the fault. The types of message indicators 72 which may be available include indications of: disturbance on the feeder; arcing disturbance on the feeder; high impedance fault; solid ground fault; and relay tripped.

The detection methodology of the preferred embodiment relies to a large extent upon the data processing which takes place in the microprocessor controller 30. The eight signals, V.sub.o, V.sub.2, V.sub.8, V.sub.h, I.sub.1, I.sub.3, I.sub.8, and I.sub.0, are sequentially selected using the eight channel multiplexer 50, following isolation by the active filter portion 26. The overall speed of the analog-to-digital conversion in the preferred embodiment is 10 kHz, which equates to 1.25 kHz for each of the eight signals. Further, a data acquisition period of 20 milliseconds is selected, which is equivalent to one period of the fundamental frequency (50 Hz) of the power system. Therefore, after the digital conversion, the microprocessor receives 200 samples (25 samples for each signal) during a given data acquisition period.

The data processing carried out in the microprocessor controller 30 follows a process of pattern forming, pattern recognition, and pattern analysis to detect a high impedance fault. This is illustrated in FIG. 10, which shows data inputs 76 resulting in pattern formations 78, which leads to pattern recognition 80 and pattern analysis 82 to enable a final decision at stage 84. The pattern formation stage 78 comprises scaling the individual samples according to the weighting factor corresponding to the frequency component signal from which the samples have been taken, and arranging the scaled samples of each signal in a respective chronologic formation. The arranged scaled samples comprise a pattern, and the formed pattern may then be stored in memory, and information about the distribution feeder can be updated by updating the patterns stored in memory. Each pattern contains information about the feeder during the corresponding data acquisition period. By continuously sampling the frequency component signals and forming and storing patterns characteristic thereof, the microprocessor controller 30 is able to continually update available information about the operating status of the feeder line. Since only a limited number of patterns can be stored in memory at any one time, past history patterns may be overwritten by more recent patterns.

FIG. 11 illustrates a block diagram of a pattern storage memory, accessible by a CPU 98, and containing four pattern storage areas 86, 88, 90 and 92, labelled pattern No. 1 to pattern No. 4. The pattern recognition stage 80 involves the comparison of patterns stored in one or more of storage sections 86 to 92 containing sequential patterns, to determine the extent of change of the frequency component signals over a period of time.

The pattern analysis stage 82 of the detection methodology involves the analysis of each pattern by examining whether signals represented exist intermittently, which is indicative of arcing.

The process of detection can be divided into several sub-processes, and a simplified flow chart 108 of the detection process is illustrated in FIG. 13. Basically, the detection system keeps updating the information about the protected feeder, by comparing newly acquired data with the preceding data, as illustrated schematically in FIG. 11. When an arcing fault occurs on the distribution feeder the detection scheme of the preferred embodiment will detect a disturbance first, and then with further processing determine whether the disturbance is a fault.

The preferred detection system can also accurately detect solid ground faults, such as line to ground faults and double line to ground faults, which are more serious in terms of the severity of the fault currents and should be disconnected immediately. These type of faults are considered first, to enable them to be detected and cleared quickly. High impedance faults have low levels of fault current, and can be tolerated by distribution systems for a few seconds.

To avoid misunderstanding, the term disturbance is chosen to describe an normal situation on a distribution feeder before the situation is confirmed as a fault. That is to say that a ground fault can be detected first as a disturbance, and then can be confirmed as either a solid ground fault or a high impedance fault.

A key process in the detection of a disturbance by the preferred embodiment is to compare two patterns formed before and during the occurrence of a high impedance fault. Data is sequentially sampled in groups of 25 samples per signal, and formed in patterns. During detection, first and second patterns are sampled and formed in a sequence, and comparison of the second pattern with the first takes place. The result of this comparison or pattern recognition determines whether the system has been disturbed.

A pattern threshold is used for measuring the difference between patterns. If the difference between the two patterns is larger than the pattern threshold, a disturbance is indicated since the change of patterns is too large. If the difference is small, then this indicates acceptable fluctuations in the normal operating condition of the feeder. The first of the patterns is then erased and the second pattern will take the position of the first pattern. Then a new pattern will be sampled and placed at the position previously occupied by the second pattern. With reference to FIG. 11, the oldest stored pattern is referred to as pattern No. 1, with sequentially newer patterns stored as patterns Nos. 2 to 4. The same routine of comparison of pattern No. 1 with pattern No. 2 then again takes place.

The determination of a value for the pattern threshold can be difficult. A disturbance can happen at any time in a pattern forming process while the microprocessor is sampling. If a disturbance occurs in the middle of pattern No. 2 as shown in FIG. 12, the differences between pattern No. 2 and pattern No. 1 or pattern No. 3 and pattern No. 2 are all very small. To tell the difference properly, the threshold should be very small. However, using such a small threshold can result in a false trip of the fault detector where the threshold is exceeded by detection of noise under normal operating conditions without a disturbance.

FIGS. 15a to 15f illustrate diagrammatically an example of a pattern comparison process for the situation illustrated in FIG. 12 where an arcing fault occurs during the data acquisition period of pattern No. 2. For the purposes of illustration, each pattern is formed from 4 frequency components (C.sub.1, C.sub.2, C.sub.3, and C.sub.4) with the height of each component representing a measure of the magnitude of the corresponding electrical signal frequency component in the distribution system during the data acquisition period for the pattern. FIGS. 15a, 15b and 15c illustrate respective patterns formed during the data acquisition periods 100, 102 and 104 of FIG. 12. FIG. 15d illustrates a threshold pattern for the particular distribution system. The threshold pattern is set to indicate a system disturbance in the event that two measured patterns differ by an amount greater than the threshold pattern. FIG. 15e illustrates the difference between patterns 2 and 1 as compared with the threshold pattern, which illustrates that no disturbance would be found in this measurement. FIG. 15f, on the other hand, illustrates the difference between pattern 3 and 1 as compared with the threshold pattern which illustrates that in this case a system disturbance is indicated. A suitable measure of the magnitude of the electrical signal frequency components, for the purposes of forming the patterns, involves tallying the number of samples of a frequency component which exceeds an individually set threshold for that frequency component, as described in greater detail hereinafter. Clearly in a pattern comparison of the form illustrated in FIG. 15, the weighting or scaling factors need not necessarily be applied to the measured pattern components, since this scaling can be accounted for in setting the threshold pattern.

To avoid mal-operation of the protection system, pattern No. 3 as shown in FIG. 12 is also examined by the microprocessor controller 30. This group of samples helps the microprocessor to make sure whether or not a disturbance has occurred. Therefore two pattern thresholds, a Low Pattern Threshold and a High Pattern Threshold are required, and three groups of samples are used for finding a disturbance. With reference to FIG. 12, the High Pattern Threshold is used in comparing the difference between pattern No. 3 and pattern No. 1, and the Low Pattern Threshold is used in comparing the difference between pattern No. 2 and pattern No. 1.

The use of two pattern thresholds results in higher precision and reliability for the detection scheme. The value for the Low Pattern Threshold can be set low enough so as to not miss a small disturbance which may occur during a high impedance fault. The value for the High Pattern Threshold can be set high enough to make sure proper reliability is achieved in not detecting non-fault related fluctuations. Very small disturbances which are not caused by a power system fault will not pass this stage.

The High Pattern Threshold is used to check a disturbance detected, by the Low Pattern Threshold. Pattern No.3 illustrated in FIG. 12 is sampled totally during a disturbance situation, which occurs during the time when pattern No. 2 is still in processing. The High Pattern Threshold which is used to measure the difference of Pattern No. 3 and Pattern No. 1 can be set relatively high and this high value is able to stop detection of very low levels and small duration noise value (less than one cycle). Therefore, the combination of the two thresholds give both high precision and reliability for this first step of detecting a disturbance.

In fact, the first step may be considered as two parts. A small noise which can overcome the Low Pattern Threshold and cannot overcome the High Pattern Threshold is named an event, and a large noise which can overcome both thresholds, is called a disturbance. Event detection and disturbance detection are the two parts of the first step of detection utilised in the preferred embodiment.

The second step of detection is the pattern analysis. As mentioned above, the use of the analysis of the intermittent nature of an arc can improve the detection of an arc. In other words, the existence of an arc can be confirmed after the pattern has been analysed.

The reason for this can be stated as follows. If the noise has been found as disturbance using multiple parameter detection and the arcing nature has been found using pattern analysis, the noise must have come from an arc. There is nothing but an arc that can create frequency component noise with the periodic characteristic where multiple frequency parameters are detected to be on and off under a certain frequency rate (100 Hz), as is discussed above.

A further difficulty encountered in detecting high impedance arcing faults is that an arc which is detected by the system is not always related to an arcing fault. The two steps of detection explained above confirm the existence of an arc in the protected feeder. However, the arc does not always occur as a result of a high impedance fault, since other non-detrimental conditions exist which lead to arc generation in power systems. For example, operations in power systems such as switching on or off of circuit breakers can also provide arcs which are detectable in the distribution system. This means that a detected arc needs to be confirmed as resulting from a fault, and this constitutes the third step of the detection scheme. An are in the power system is called an arcing disturbance before it is confirmed by the detection system.

The method used to tell the difference between an arcing fault and an arcing disturbance is based on introducing a time delay. Normally, an arc caused by normal system operations lasts only a short time because modern circuit breakers in particular work very fast. For example, an arc between two points in a circuit breaker may last less than 0.1 second. In contrast, an arc from an arcing fault will last much longer than that. Further, an arc on a high impedance object may last more than several seconds, even minutes before it develops into a hazard such as bush fire. There is no need for protection action if an arcing fault is able to quickly self-clear without circuit breaking intervention. A short time delay in the detection scheme can also be tolerated by power systems because of the low magnitudes of arcing high impedance fault currents. Arcing faults cause low level current, which does not ordinarily affect power system stability. Finally, the fault can be detected by the ground fault detection relays if the arcing current is high.

The primary idea of confirming a high impedance fault is simple. By processing more samples and forming a new pattern after a time delay, the microprocessor can compare the updated pattern with pattern No. 1 which is the only one sampled before the fault. The previously discussed method of pattern recognition is used for the comparison, and pattern analysis is used to analyse the new pattern. The result of these comparisons can answer whether the arcing fault still exists on the feeder after the introduced time delay. The trip instruction may be issued if the answer is confirmed.

However, the unstable nature of an arcing fault makes the confirmation process difficult. A large arcing current may change to small or even stop for several cycles because of the wind or the movement of arcing points. If samples are introduced at this time, the confirmation of the fault will fail. Of course, for the situation where the arc can stop by itself and remain off permanently, no tripping is the right choice. For the situation where the arc persists with an unstable burning nature a failure of confirmation should be avoided.

Confirmation of a high impedance fault becomes complete when two new patterns can be formed and used to confirm the existence of an arcing fault. They will not take more time for the confirmation because the microprocessor has to wait during the time delay. So the whole time delay can be divided into two, at the end of the first half of the delay the system performs one confirmation operation, and another confirmation operation is performed after the whole time delay. A suitable total time delay is 2 seconds.

Solid ground faults such as line to ground and double line to ground faults have relatively large fault currents compared to high impedance fault currents. They can be detected easily since zero sequence current Io will increase high enough to be easily detected. Solid ground faults can be detected relatively quickly. In order to achieve this, checking the occurrence of a solid ground fault has the highest priority in the detection scheme. In other words, in the microprocessor procedures the detection of occurrence of a solid ground fault is designated just after each new pattern is formed. A solid ground fault therefore can be detected quickly. To produce each pattern, the microprocessor needs 0.02 second. In a normal detection routine, the microprocessor samples continuously without a time delay except a very little time for performing the instructions. The time for performing the instructions takes a few millisecond which can be neglected compared to 0.02 seconds. Therefore a solid ground fault can be noticed by the microprocessor in about 0.02 seconds. The time for the system to trip depends on the detection scheme used.

Solid ground faults which develop from arcing faults can also be detected easily. In some cases, a fault starts with an arc which initially presents a high impedance fault and then it develops into a solid ground fault. By continuously checking the most up to date patterns, this kind of fault can be detected. This is an additional feature incorporated in the preferred embodiment high impedance fault detection scheme.

The time to respond to a developing ground fault varies from 0.02 second onwards. If the development of a fault has finished before the program starts the arcing fault confirmation routine, the fault will be detected in about 0.02 second. If the development finishes after that, the fault will be detected quicker, because during the confirmation routine the microprocessor has more free time to handle this.

In terms of time delay, the effect of solid ground fault detection on high impedance fault detection can be neglected. During high impedance faults, the final trip instruction will not be issued until four or five patterns are sampled. After each pattern is sampled, a quick check for the solid ground fault takes place. It takes only several milliseconds for each quick check, and so has almost no effect on a high impedance fault detection in terms of time delay.

A simplified methodology for performing the pattern recognition portion of the detection system involves combining the weighted data forming a pattern into a single weighted sum of frequency components. Individual thresholds T.sub.i may be set for each frequency components to isolate the signal samples which exceed the relevant threshold. The number of samples of a particular frequency component signal which exceed the given threshold during a data acquisition period may be utilised by the microprocessor controller 30 as a measure of the magnitude of that frequency component. If the number of samples of each signal exceeding the individual thresholds is then scaled by a weighting factor and combined with the weighted samples of the other frequency components then a single quantity is obtained which is characteristic of the weighted magnitudes of the examined frequency component signals. This single quantity may be easily compared with quantities obtained during other data acquisition periods to perform the pattern recognition function described above. The summing and scaling of the predetermined frequency components may be conveniently carried out according to the following formula: ##EQU2## Where: X is the weighted sum of the frequency components

n is the number of frequency components

W.sub.i is the scaling factor for the frequency components

m is the total number of samples during a data acquisition period

S.sub.ij is the sample magnitude for the frequency components

T.sub.i is the individual threshold for each frequency component.

With reference then to flow chart 108 illustrated in FIG. 13, the detection methodology carried out by the microprocessor controller 30 begins at step 110, and at step 112 the frequency component signals are sampled over a data acquisition period. The data obtained may then be stored in memory either as individual weighted frequency component signals, or as a weighted sum of components as described above. A system disturbance is detected at step 114, either by comparison of the pattern data obtained at step 112 with pre-stored standard pattern data, or by comparison with pattern data obtained in the immediately preceding data acquisition period. If the magnitude of the zero sequence signals indicates a solid ground fault at step 116, then the procedure continues to step 118 where the solid ground fault may be confirmed and action taken to isolate the fault. If a solid ground fault is not indicated the procedure continues to step 120 where pattern analysis is carried out on the data obtained at step 112 to confirm whether the system disturbance detected at step 114 is in fact attributable to an electrical arc. If an arc is confirmed at step 120 then data samples are again taken during a further data acquisition period (step 122). A comparison of the pattern of data obtained at step 112 and that obtained at step 122 may then confirm an arcing fault (step 124) in which case appropriate message signals and trip signals are issued at step 126.

Flowchart 128 illustrated in FIG. 14 represents a slightly more complex detection procedure which utilises the Low Pattern and High Pattern thresholds. Steps 130 and 132 represent the initialisation steps of the procedure where parameters such as the individual signal thresholds are retrieved and weight or scaling factors for each signal are determined. At step 134 it is determined whether an event has taken place in the distribution system by comparison of two sequential data acquisition period patterns with the low pattern threshold. If an event is determined (step 136) the procedure continues to step 138, or else the procedure continues to monitor for an event at step 134. Once an event has been detected (step 136) the occurrence of a solid ground fault is checked at step 138. In the event of a solid ground fault a time delay is introduced (step 158) and the ground fault is confirmed or denied at step 160.

If the detected event (step 136) does not result from a solid ground fault then further samples are obtained (step 140) to enable detection of an arc disturbance in steps 142 to 146. Having confirmed that the disturbance is due to an electrical arc a fault condition is confirmed at step 150, which results in the generation of relevant trip signals to isolate the fault from the electrical distribution system (step 152).

In appendix A there is shown an example of weights, thresholds and amplitude adjustment scales which have been successful in tests on a three wire three phase distribution system. It must be born in mind, however, that the weights indicated are by way of example only, and can be adjusted according to experience from the individual power system. Ranges of relative component weights which have proved successful in practice are as follows:

In addition, by changing the weight for each signal, the detection scheme has the ability of learning. The weight values can be set for each protection relay initially according to the previous experience of the utility including data collected during previous arcing faults, and can be readjusted periodically. This ability of learning provides an opportunity to improve the accuracy of the detection. The weight can be adjusted to suit a given utility condition without the need for changing the hardware, since such changes can be implemented in software.

Tabulated and described herein below are the results of a field test of arcing fault detection apparatus utilising the simplified methodology of the preferred embodiment of the invention. The first step of the test involved monitoring and sampling the prescribed eight signal components on an electrical feeder line during normal operation. An electrical fault condition was then induced in the feeder line by severing a cable and laying the severed cable end on an asphalt surface, this taking place with the electrical supply disconnected. Monitoring and sampling was again commenced, and continued whilst the electrical supply was reconnected to induce an arcing fault condition. Monitoring, sampling and processing of the signal components continued during the arcing fault until an isolating relay was tripped following confirmation of an arcing fault by the detection apparatus.

Shown below are the maximum voltage and current levels of the signal components before the fault condition, and the samples thereof recorded during a corresponding data acquisition period.

During the arcing fault, before the isolating relay was tripped, the maximum voltage and current levels of the signal components were recorded as follows:

The samples recorded during a data acquisition period occurring during the fault are shown below:

The relation between the analog lues of voltage and current for the signal components and the digitised sample values are in accordance with the tabulated conversion values shown in Appendix A.

By way of example, if the individual threshold levels for the signal components are considered as zero, the pattern sums X for the above sample data may be easily calculated. In effect the pattern sum X for sample data during a given data acquisition pattern may be determined by: i) summing the number of samples for each signal component in which the sample value is greater than the corresponding individual threshold; ii) multiplying each sum resulting from i) by its corresponding weight value; and iii) adding the weighted sums to determine the overall pattern sum X. This process is illustrated step-by-step below for the test data both before and during the arcing fault, with zero individual thresholds.

Clearly, using this strategy and the pattern sum threshold from Appendix A of 40 (Hex), a fault condition would be detected since the difference between the pattern sums before and during the fault is greater than the threshold, ie: ##EQU3##

In practice it has been found that a threshold level of 10 (Hex) for the individual signal components yields accurate fault detection whilst substantially eliminating the effects of ordinary voltage and current fluctuations on the detection scheme. It must be kept in mind, however, that the individual threshold levels for the signal components must be balanced against the overall pattern sum threshold, since each affects the sensitivity of the fault detector. This is clearly demonstrated by calculating the pattern sums for the above sample data before and during a fault using an individual threshold level of 10 (Hex) for each component, which yields:

Thus, to detect this particular fault using an individual threshold level of 10 (Hex), it would be necessary to have a pattern sum threshold which is less than 23 (Hex) (i.e. less than X.sub.d -X.sub.b).

Appendix B shows a listing of computer software program code in assembly language suitable for a Motorola MC6802 microprocessor, which has been developed to implement the functions of the described embodiment which are controllable by the microprocessor control unit 30.

Although the foregoing description has been explained in relation to a single phase arcing fault, such as a single phase line contacting a tree or the like, the technique and apparatus described herein is in fact applicable to a number of applications. For example, other arcing faults and phenomena can be detected, such as partial discharge through high impedance medium resulting from leakage currents in aging or failing insulation material. Furthermore, high energy faults such as two phase and three phase line to line faults can be detected where these result in arcing between the conductors. Certain types of equipment used on the electrical distribution system can also be detected, such as arc furnaces and are welders, which can be useful for the recording purposes of the electrical supplier or for protecting sensitive electrical equipment on the distribution system from any adverse effects of the arcing equipment.

The foregoing has been put forward by way of example only, and many modifications will be apparent to those skilled in the art without departing from the spirit and scope of the invention as defined in the claims appended hereto. ##SPC1##