WO2008069979A1

WO2008069979A1 - An improved method and apparatus for detecting the high impedance fault during close-in-to fault and auto-reclose situations

Info

Publication number: WO2008069979A1
Application number: PCT/US2007/024663
Authority: WO
Inventors: Ratan Das; Mohamed Y Haj-Maharsi; John M. Peterson
Original assignee: Abb Technology Ag
Priority date: 2006-12-01
Filing date: 2007-11-30
Publication date: 2008-06-12

Abstract

A fault detection system is disclosed that connects to an electrical power line and detects high impedance faults. The system includes high impedance fault detection system or systems that are operable to detect a high impedance fault from a signal taken from the electrical power line. The high impedance fault detection system employs a first threshold value for a first predetermined period of time and a second threshold value after the first predetermined period of time.

Description

AN IMPROVED METHOD AND APPARATUS FOR DETECTING THE HIGH IMPEDANCE FAULT DURING CLOSE-IN-TO FAULT AND AUTO-RECLOSE

SITUATIONS

Technical Field

The present invention relates to a method and an apparatus for improving the detection of high impedance faults in electrical power systems, in particular during close-in- to fault and auto-reclose situations.

Cross-Reference Related Applications

This application claims priority to provisional application No. 60/872133 filed on December 1, 2006 and entitled An Improved Method and Apparatus for Detecting the High Impedance Fault During Close-In-To Fault and Auto-Reclose, the contents of which are incorporated by reference in their entirety.

Background

High impedance faults result when an energized primary conductor comes in contact with a quasi-insulating object, such as a tree, a structure or equipment, a pole cross-arm, or falls to the ground. Typically, a high impedance fault exhibits arcing and flashing at the point of contact. The significance of these hard to detect faults is that they may represent safety problems as well as a risk of arcing ignition of fires. As such, high impedance fault detection has been a major concern of protective relaying for a long time.

Protective relays are usually designed to protect equipment (line, transformer, etc.) from damage by isolating the equipment during high current conditions. High impedance faults, are typically found on distribution circuits, results in very little, if any, current. High impedance faults do not pose a threat to equipment and by their nature they can not be detected with conventional over-current devices. Nonetheless, the dangers of a downed conductor are obvious to all. Possibility of fire, property damage, and someone coming into contact with the live conductor are some of the major concerns.

Providing a comprehensive solution for high impedance faults is a difficult issue. For example, tripping the breaker following high impedance fault detection is not a clear cut choice. While the high impedance fault is a danger, tripping the feeder unnecessarily will create new problems by de-energizing homes, traffic signals, offices, etc. The effects of incorrectly determining a high impedance fault will have adverse ramifications, and the goal of any HIF detection method is to provide very high security with a reasonable degree of dependability, which requirements are always in contradiction and there is a delicate balance between them.

Some HIF detection techniques have been proposed which are based on the comparison between the pre-fault energy and the energy during fault.

The pre-fault energy during close-in-to fault or auto-reclose situations is practically zero. To prevent possible misdetections by the implemented techniques, a high secure value is used as pre-fault energy in place of low or zero values. This secure value may be too high in most cases but there is not one value that is suited for all feeders. Further, a feeder may experience a HIF during re-closing into a low-impedance fault if the deionization of the fault environment has not been completed during the re-closing window. Accordingly, detection of HIF conditions during re-closing is very important like the detection of HIF conditions during close-in-to fault situations.

Summary of the Invention

According to one embodiment, a method for detecting a high impedance fault in an electrical power line after the opening and closing of a breaker is provided. The method includes providing a plurality of high impedance fault detection means, providing a decision means, using the plurality of high impedance fault detection means to make a plurality of independent determinations from a signal taken from said power line. The plurality of high impedance fault detection means employ a first threshold value for a first predetermined period of time after the close of the breaker and a second threshold value after the first predetermined period of time. Outputs are generated representative of the independent determinations, and decision means are used to determine whether said high impedance fault has occurred.

According to another embodiment, a fault detection system is connected to an electrical power line for detecting a high impedance fault after the opening and closing of a breaker. The high impedance fault detection system includes a plurality of high impedance fault detection systems operable to respectively make a plurality of independent determinations from a signal taken from the electrical power line. The plurality of high impedance fault detection systems employ a first threshold value for a first predetermined period of time after the close of the breaker and a second threshold value after the first predetermined period of time, to respectively generate outputs representative of the independent determinations. Decision means are connected to the high impedance fault detection systems for determining whether the high impedance fault has occurred.

According to yet another embodiment, a fault detection system is connected to an electrical power line for detecting a high impedance fault after the opening and closing of a breaker. The high impedance fault detection system includes a high impedance fault detection system operable to detect a high impedance fault from a signal taken from the electrical power line. The high impedance fault detection system employs a first threshold value for a first predetermined period of time and a second threshold value after the first predetermined period of time. The second threshold value is dynamically determined by sampling the signal from the electrical power line, and the first predetermined period of time is at least the minimum amount of time required to acquire sufficient samples to dynamically determine the second threshold value.

Brief Description of the Drawings

Fig. 1 shows a schematic diagram of an electrical power distribution system. Fig. 2 shows a high impedance fault detection system. Fig. 2a shows a system in which the fault detection system of Fig. 2 is used. Figs 3 and 3a show arrangements for the filters of Fig. 2a.

Figs. 4, 5 and 6 show the characteristics of the notch filters for the fifth, sixth and seventh harmonics of the power line frequency.

Fig. 7 is a flowchart showing an HIF detection method using a plurality of threshold determining methods.

Fig. 8 is a flowchart showing an exemplary wavelet based HIF detection application. Fig. 9 is a flowchart showing an exemplary higher order statistics based HIF detection system.

Fig. 10 is a flowchart showing neural network based HIF detection application. Fig. 1 1 is a block diagram for the dynamic threshold HIF detection technique.

Detailed Description of the Preferred Embodiment(s)

In this invention, a method and apparatus to improve a proper detection of high impedance faults during close-into-fault or auto-reclose situations are provided. In the proposed approach security has been given a higher priority over the dependability.

Figure 1 shows a schematic diagram of an electrical power distribution system having an electrical power distribution line 10 and a high impedance detection system 12. The solid vertical bars 16 in figure 1 are bus bars and represent the interconnection of multiple distribution lines. The high impedance detection system 12 include one or more individual high impedance fault detection systems 22, 24, 26 which are shown in figure 2. Also shown in figure 1 are the potential transformer PT and the current transformer CT which provide the typical analog inputs for a protective relay which is operatively coupled to an associated circuit breaker 1 1.

These individual high impedance fault detection systems have individual algorithms 18 for individually detecting high impedance faults as described in patent US 7,069,1 16 ('" 1 16 patent") whose content is hereby incorporated by reference. These algorithms can use, for example, wavelets, higher order statistics, neural networks, and the like to identify the presence of high impedance fault independently of each of the other system algorithms. The individual high impedance fault detection algorithms can each have a different confidence level. A fault is identified as a high impedance fault once it is detected independently by the algorithms and processed through a decision logic.

Figure 2 shows an exemplary composite high impedance fault detection system 12 including a higher order statistics based high impedance fault detection system 20 identified in figure 2 as a 2^nd order statistical system 22, a wavelet based high impedance detection system 24, and a neutral network based high impedance detection system 26. An input connection 28 labeled "Acquisition" in figure 2 and an output connection 30 labeled "Detection decision" in figure 2 are provided for communicating an electrical signal between the electrical power distribution system and the high impedance fault detection systems 22, 24, 26.

For example, the input connection 28 receives an electrical signal from a sensing device coupled to the electrical power distribution line. The sensing device can include any suitable sensing device, such as the current transformer shown in figure 1. The output of acquisition 28 is processed through data filtering means 29 which provides band limited and filtered signals to each individual high impedance fault detection systems 22, 24, 26. The filtering means 29 are preferably software filters and are implemented on the CPU board 34 illustrated in figure 2a.

As illustrated in figure 3, the filtering means 29 comprises a first pass band filter 29a, and one or more additional notch filters for different harmonic components. The band pass filter range for the sampled phase and/or ground current signal(s) is for example 297- 430 Hz for 60 Hz power systems, that is from slightly below the frequency of the fifth harmonic to slightly above the frequency of the seventh harmonic. The band pass filter range is adjusted accordingly for the 50 Hz systems. In the embodiments illustrated, the filtering means 29 preferably comprise three different notch filters, namely a sixth harmonic notch filter 29b for sixth harmonic components, a seventh harmonic notch filter 29c for seventh harmonic components, and a fifth harmonic notch filter 29d for fifth harmonic components. These filters can be used independently or as a group with all possible combinations among them.

In particular, in the embodiment illustrated in figure 3 the sampled signal is filtered in cascade first by the band pass filter 29a, then by sixth harmonic notch filter 29b and then by the seventh harmonic notch filter 29c to generate a Signal 1. The band pass filter 29a filters the sampled signal to remove all frequencies outside the filter range, the sixth harmonic filter 29b removes the sixth harmonic component from the band pass signal and the seventh harmonic filter removes the seventh harmonic component from the band pass signal without the sixth harmonic component. Therefore, Signal 1 only contains frequencies including the fifth harmonic that are within the band pass filter range without the sixth and seventh harmonic component. The fifth harmonic notch filter 29d removes the fifth harmonic component from Signal 1 to generate Signal 2.

Once filtered, the signals 1 and 2 are fed to the HIF systems 22, 24, 26 for proper processing through the algorithms 18 as will be better described hereinafter. As is described in the incorporated herein by reference ' 1 16 Patent, the wavelet high impedance fault detection system 24 of Fig. 2 makes use of the fifth harmonic of the band limited and filtered signal from means 29 of Fig. 2 whereas the fifth harmonic is not required for the other high impedance fault detection systems 22 and 26 shown in Fig. 2. Thus Signal 1 is fed to at least the wavelet system 24.

Alternatively, as shown in figure 3a, the filters can be arranged in series one after the other with the fifth harmonic notch filter 29d which receives at its input the sampled data 28, the sixth harmonic notch filter 29b connected to the fifth harmonic notch filter 29d, and the seventh harmonic notch filter 29c connected between the sixth harmonic notch filter 29b and the band pass filter 29a. Other configurations may be adopted as well.

The characteristics of the notch filters 29b, 29c, and 29d are shown in figure 4, figure 5 and figure 6 respectively. These filters are designed to have sufficient attenuation of the related harmonic frequency for a system frequency variation up to + 3%.

The presence of the filter means 29, and in particular of the notch filters 29b, 29c, and 29d, allows to reduce, if not completely eliminate, possible mis-operation of the HIF detection system which may be caused by the presence of large time-varying load harmonic components which are within the band pass filter range used.

Figure 7 is a flow diagram illustrating the improved method for detecting HIF during close-into-fault and auto-reclose situations according to the present invention. As it will be better described hereinafter, the method according to the present invention uses energy values from two different memory locations 1 or 2 as pre-fault values. These two memory locations 1 and 2 can be implemented on the CPU board 34 shown in figure 2a. The energy value in memory location 2 is very high and accordingly very secure. The method uses energy values from location 1 as the first option. This energy value is based on the load value in the feeder within a given time window indicated as Tl hours in figure 7. In one embodiment, the value is chosen as 18 hours. Load energy values are continuously calculated and stored in memory 1 by the microprocessor associated of the CPU board 34. This time window appears to be a reasonable compromise between the security and dependability, however different values can be defined according to the applications or specific needs. It is assumed that load characteristics will not change significantly during the time window Tl . However, to further improve the security, a variable called 'Factor', shown in Figure 7 is used to enhance the pre-fault energy values which accommodates some degree of possible variations in the load during the close-into-fault and auto-reclose situations. The time window T2, which is for example expressed in seconds, depends on the window used in a dynamic energy calculation as the one described in US patent 7,085,659 (the '"659 patent") whose content is hereby incorporated by reference. In one embodiment, the value of the time window T2 is selected for example as 80 sec.

In practice, as shown in figure 7, when the switching component, e.g. the circuit breaker 11 is closed at 100 thus restoring the flow of power in the electrical power system, the microprocessor unit checks at a first step 101 if the breaker remained open for a period of time longer than a first predetermined interval of time Tl . When closing of the breaker is detected, also a timer is started by the microprocessor. If at step 101 it is verified that the breaker remained open for less time than the first time interval Tl , then at step 102 it is checked if a load energy level value is available in a first memory location 1. If this load energy level value is not available in a first memory location 1, then at step 103 a predetermined energy value which constitutes a fixed threshold is recovered from a second memory location 2 and is used in processing HIF algorithms which are implemented on HIF detections systems. These algorithms can be the algorithms 18 implemented on HIF detection systems 22, 24, 26 according to the solutions described in the ' 1 16 patent. If an HIF condition is detected at 103, a detection signal 30 is output by the logic 32.

When instead at step 101 it is verified that the breaker 11 remained open more for a longer time than the first time window Tl , the algorithm at the base of the method according to the invention passes directly at step 103 and operates exactly as above described.

The monitoring using HIF algorithms with the fixed threshold of memory location 2 continues until, at step 104 it is verified that the time elapsed from the instant the timer is started to the actual instant is greater than a second predetermined interval of time T2. If this condition is not verified, then steps 103 and 104 are repeated again and again until it is actually verified that the time elapsed from the instant the timer is started to the actual instant is greater than the second predetermined interval of time T2. Thus, the HIF algorithms using the fixed threshold of memory location 2 monitor the energy level until time T2 from closing the breaker. In other words, steps 103 and 104 loop until sufficient time that the FIFO buffer of the dynamic thresholds is filled.

If the requirement of 104 is satisfied, at step 105, algorithms with dynamic thresholds are carried out according to the solution disclosed in the '659 patent. If at step 106 an HIF condition is detected, then at step 107 the energy level update for each signal and algorithm is frozen in order to prevent storing load energy levels when a HIF condition is detected. If a HIF condition is detected, a detection signal 30 is output by the logic 32. If instead an HIF condition is not detected at step 106, then at step 108 a value calculated by multiplying the load energy level for a security variable is compared with the energy level value stored in the first memory location 1. In one exemplary embodiment the value of 'factor' is chosen as 2. This value appears to be a reasonable compromise between the security and dependability, however different values can be defined according to the applications or specific needs. If the value calculated by this multiplication is not greater than the energy level in the memory location 1, then the memory location 1 is not updated and the method repeats step 105 with dynamic thresholding and proceeds therefrom as previously described. The energy calculated in step 108 is always used for step 105 to update the FIFO buffer of threshold values. If instead the value calculated at step 108 is greater than the energy level in the memory location 1, at step 109 the load energy level value stored in memory location 1 is updated with the value calculated at step 108 for future use at step 102 and 103a. This calculated value is then used at step 105 to update the FIFO buffer and the method proceeds from this step as previously described. When at step 102 the load energy level value is available in the first memory location 1 , the algorithm would proceed instead at step 103a equivalent to step 103 but in this case the energy value used as the threshold input in processing HIF algorithms is that stored in memory location 1. Also in this case, the algorithms used may be those algorithms 18 described for example in the ' 116 patent. Under this situation, at step 106 (in figure 7 there are two boxes 106 instead of one for the sake of clarity of illustration) it is verified whether a HIF condition is detected. If yes, at step 107 the energy level update for each signal and algorithm is frozen with the actual value and a detection signal 30 is output by the logic 32. If the HIF condition is not detected, at step 104 the condition previously described is verified (in figure 7 there are also two boxes 104 instead of one for the sake of clarity of illustration). If at step 104 it is verified that the time elapsed from the instant the timer is started to the actual instant is not greater than the second predetermined interval of time T2, the algorithm returns to step 103a and proceeds accordingly as before described. If instead the time elapsed from the instant the timer is started to the actual instant is greater than the second predetermined interval of time T2 (and accordingly the FIFO buffer for the dynamic thresholding is full), the previously described step 108 is carried out and the method proceeds accordingly as before illustrated.

As above indicated, at step 103 and equivalent step 103a suitable HIF algorithms 18 implemented in detection systems such as the systems 22, 24 26 are appropriately carried out, while at step 105 the HIF algorithms with dynamic thresholding are carried out in an exemplary way that will be described hereinafter. Thus, for the first time period T2 after a breaker closes, HIF algorithms using a fixed (very high) threshold is used or, if available HIF algorithms using a threshold from memory 1 is used. After the lapsing of time T2, the HIF algorithms are implemented using dynamic thresholding.

As illustrated in figure 2a, after the data 28 are acquired, for example by means of the combination of the potential transformer PT and the current transformer CT shown in Figure 1 and then filtered by hardware filters 21. After proper processing, the signals filtered by the filter means 29 which are preferably operatively associated with and implemented on the CPU board 34 are supplied to the high impedance systems 22, 24, 26. As shown in figure 2, each individual high impedance fault detection system 22, 24, 26 includes a logical output that is communicated to the composite high impedance fault detection system shown in Figure 2 as "Decision Logic" 32 which determines whether a high impedance fault has occurred. The composite high impedance fault detection system detects and identifies a fault as a high impedance fault once it determines that at least one, preferably at least two individual high impedance fault detection systems 22, 24, 26 have independently detected a high impedance fault. This composite feature provides increased security against false identification while improving the probability of detecting all high impedance faults. Each high impedance fault detection system 22, 24, 26 and its associated algorithm 18, as well as the composite algorithm are discussed in detail below.

The output connection, that is "Detection decision" 30 of Decision Logic 32 provides the logical output from each of the individual wavelet based high impedance detection systems, that is, the higher order statistics based high impedance detection system 22, the wavelet based high impedance detection system 24 and the neural network based high impedance detection system 26, to the composite high impedance detection system.

The higher order statistics based high impedance detection system 22, the wavelet based high impedance detection system 24 and the neural network based high impedance detection system 26 and the decision logic 32 are stored in memory and implemented in a microprocessor which is also used for implementing non-HIF detection algorithms such as protection, other than HIF detection, and control algorithms and if desirable metering and/or monitoring algorithms. Thus in the present invention, one microprocessor is used for implementing both non-HIF detection and HIF detection algorithms.

One example of how to implement the HIF algorithms using a digital signal processor (DSP) is described in US patent application Ser. No. 1 1/081,042 filed on March 15, 2005 whose content is hereby incorporated by reference.

As shown in better detail in figure 2a, the filtered signals outputted by the filtering means 29 are provided to a multiplexer 23. The output of multiplexer 23 is connected by an analog to digital converter 25 to the input of a digital signal processor 27. The embodiment shown in figure 2a also includes a memory 33 and a CPU board 34 which includes a microprocessor 34a, a random access memory 34b and a read only memory 34c. As was described above in connection with figure 2, each of the individual high impedance fault detection systems 22, 24, 26 shown in that figure are implemented in microprocessor 34a. Also as was described above in connection with Figure 2, microprocessor 34a is also used for implementing non-HIF detection algorithms such as protection, other than HIF detection, and control algorithms and if desirable metering and/or monitoring algorithms. The output of CPU board 34, which is an indication that a high impedance fault or a non- high impedance fault condition was determined is connected to alarming 36.

Figure 8 is a flowchart showing an exemplary wavelet based HIF detection application. After the data is acquired in 50, it is filtered in 52, and then, as is described in detail below, it is decomposed in separate high and low pass wavelet decomposition filters in 54. The energy is then calculated in 56 and the calculated energy is compared to a threshold in 58 to determine if a HlF has occurred.

The following is an exemplary application of high impedance fault detection using a wavelet based high impedance fault detection system. The continuous wavelet transform of

>^•(') is

where, the wavelet is φ(t), p is the position and s is the scale.

The position argument keeps track of the temporal change in current harmonics which is essential to HIF detection and the scale change keeps track of bands of frequencies of the current load. Both position and scale are continuous, therefore the above transform is not suited for computation. A discrete version of the transform is needed which is given by,

C _mn = ∑ r d ) _φ (^-) (2) k V 2 J where, k, m and n are all integers.

The above transform is implemented by multi resolution analysis where the signal is decomposed into a low pass and a high pass component via two separate low pass and high pass filters known as wavelet decomposition filters. After filtering, both low pass and high pass signals are down sampled by a factor of 2. The high pass signal component corresponds to the first detail look of the signal. The second detail look can be obtained by further decomposition of the current low pass signal into two new low pass and high pass components. The third, fourth, etc. detail signals can be obtained by further decomposition of subsequent low pass components.

The original signal can be reconstructed with minimal error from its low pass and high pass components in a reverse pyramidal manner. It is in these high pass components where distinct HIF features can be located and distinguished from signatures of other nonlinear loads of transient and bursty nature. The decomposition filters are associated with the type of mother wavelet used.

Most of the exemplary tests of this technique were conducted using the Daubechies- 4 wavelet which is not a very smooth wavelet but requires less computation time. Use of other wavelets or other Daubechies wavelets did not show any noticeable change in performance nor in the threshold parameters used. The exemplary HIF detection algorithm developed for the wavelet based system examines overlapping windows of the current at different scales and details via a wavelet transform. Although proper HIF detection can be accomplished using more than a single scale, experimental testing indicated that the energy component of the seventh detail signal carries the most significant HIF information that is more distinguishable from other normal arcing loads or normal nonlinear loads. The additional preprocessing needed is a FFT to render the current with all its random delay components position insensitive.

Thus, the preferred algorithm relies on evaluating the energy of the seventh detail signal of the magnitude of the FFT of a current. That energy is compared to a threshold and to the energy of the previous data segment. The combined decision results in a fault/no fault determination. This detection scheme delivers about 80% detection with about a 0.5% false alarm rate in the absence of arc welding loads. If the HIF attenuation parameters were lower limited to 0.1 (i.e. typically high impedance fault detection systems are not interested in detecting very weak currents), the detection rate increases to about 95% with about a 0.1% false alarm rate. The detection performance drops to about 65% in the presence of arc welding signals and without considering any lower limits on attenuation. The false alarm rate remains under about 1%.

Figure 9 is a flowchart showing an exemplary higher order statistics based HIF detection system. The data is acquired in 50, it is filtered in 52. The data acquisition and filtering in this application are both the same as the data acquisition and filtering described for the wavelet based HIF detection system of figure 8 and thus have in figure 9 the same reference numerals as is used in figure 8 for those functions. The energy is then calculated in 60 and the calculated energy is compared to a threshold in 62 to determine if a HIF has occurred.

An exemplary detection system and algorithm based on examining the higher order statistical features of normal currents has been developed and tested, as discussed below. Higher order spectra, namely the bispectrum and trispectrum are traditionally recognized as important feature extraction mechanisms that are associated with the third and fourth order cumulants of random signals. The bispectrum and the trispectrum are by definition the two dimensional and three dimensional Fourier transform of the third and fourth order cumulants defined as,

C 2 (m, n) = E{r (t) r (t + m) r (t + n)} (3)

C 3 (m, n, k) = E{r (t) r (t + m) r (t + n) r (t + k)} (4) where E stands for the expected value.

The exemplary algorithm implemented in this study is due in part to Tugnait (see, J. Tugnait, "Detection of Random Signals by Integrated Polyspectral Analysis", IEEE Transactions on Signal processing, Vol. 44, No. 8, pp. 2102-2108, August 1996) and utilizes the integrated polyspectra of single-phase current loads. This reference in incorporated herein by reference in its entirety.

The detector is developed such that a detection decision is made either using second order statistics at a preliminary stage or using third and fourth order statistics at an additional stage. The basic concept is as follows: what is the achievable detection decision assuming accessibility to second, third, and fourth order statistics for a given set of data and a fixed false alarm rate. First, it is determined whether a fault exists using only second order statistics. If the detection cannot be made, an alternative test based on third and fourth order cumulants is triggered. Both tests combined are designed such that the probability of false alarm is fixed and predetermined by the system operator. Clearly, this detector uses additional information beyond energy signatures.

Preferably, this detector relies on all current spectra including the in-between harmonics as generated by the pre-processing filter described earlier. The HIF detector is itemized as follows:

I. Declare a fault, if s_ecj > τ_a

Where, the signature s_e({ denotes the second order statistics of the data r(t) and T_a is the threshold. s_ecj is defined as,

where, σ% is the variance of r(t) given Hypothesis H(O) which is a no fault situation. The threshold T _a is chosen such that,

where, χ_N ² (o) denotes the non-centered chi-squared distribution of N degrees of freedom.

The parameters a_s and a^ are set by the designer such that, a_s + (\ -a_s )a_h = a (7)

where, a is the predetermined probability of false alarm. II. If a detection cannot be made with the previous test, then the following step is used. Declare a fault, if,

sh > ^τa_h

where, the signature _^■ _/, denotes the third order statistics of the data r(t) and τ_Oh is the threshold.

The threshold T_Uh is chosen such that,

where, (p) denotes the non-centered chi-squared distribution of N degrees of freedom. The parameters a_s and a_/, are set by the designer such that,

a_s +(\ -a_s )a_h = α (9)

where, a is the predetermined probability of false alarm.

The data of length N is divided into L segments each of length N _B . The third order statistics s_h is a scalar and defined as,

** = ²∑ Yf K XDh m O Y^λV_h (ω_m ), (10)

OT = I

The vector V_f, is defined as,

K_A(©_m ) = [Re{s_C2rl.(«)}lin{s_C2rl.(m))Re{s_C3rl.(«)}Re{s_C3rr (mjf^"

(H)

The transpose symbol used is T and ω_m = JΞL vvith m = 1,2, ,L .

N B

Thus, ω_m represents all the spectral components of the recorded current. The real and imaginary components are denoted by Re and Im respectively. The inverted matrix D_/, _{m Q} used in the example above is defined as a diagonal matrix with elements representing the integrated polyspectra of the no fault signal. K ))]

(12)

The integrated bispectral and trispectral components are defined as,

where, C _/r = FFT {c_/r } and the cumulants are defined as,

ci_r (14)

Finally, Λ (α>£ ) is the Fourier Transform r(t).

Figure 10 is a flowchart showing neural network based HIF detection application. The data is acquired in 50, it is filtered in 52. The data acquisition and filtering in this application are both the same as the data acquisition and filtering described for the wavelet based HIF detection system of figure 8 and thus have in figure 10 the same reference numerals as is used in figure 8 for those functions. The samples are transformed in 64 using a fast Fourier transform (FFT) which is used only in the second neural network embodiment described below, and then mapped into the HIF plane in 66 using the neural network algorithm and compared to a threshold in 68 to determine if a HIF has occurred.

The following is an exemplary application of high impedance fault detection using a neural network based high impedance fault detection system. Artificial Neural Networks (ANNs) have been successfully used in many applications to solve complex classification problems due to their ability to create non-linear decision boundaries. The most common and flexible neural network is the multi-layer perceptron (MLP) which is constructed from a series of neurons.

One embodiment of a neural network design used the spectrum of the 3-cycle window of data. The magnitude of the FFT of the 1000 samples was truncated at the 13th harmonic. This resulted in a reduction to only 40 input nodes for the neural network. This network had fewer weights and biases and could be trained almost an order of magnitude faster. The best results occurred when 30 nodes were used in the hidden layer. The network was trained with 600 cases and had a sum-squared error of 1 1.8 (8 missed detections and 4 false alarms). Generalization testing on 3600 new inputs resulted in about an 86.06% detection rate with about a 17.06% false alarm rate. The increased performance of this network over the previous network is likely due to the invariance of the frequency spectrum to phase shifts. These performance figures are once again based upon using about 0.5 as the output threshold for indicating a detected HIF. An attempt was made to reduce the false alarm rate by increasing the output threshold to about 0.75. This resulted in about a 83.7% detection rate with about a 14.8% false alarm rate. Increasing the threshold to about 0.95 resulted in about a 77.7% detection rate and about a 1 1.8% false alarm rate.

Another exemplary network architecture was a combination of the two previous networks operating in parallel. If the output of both networks was greater than 0.5, then a positive HIF decision was indicated. A decision that no HIF was present was made if the output of both networks was less than 0.5. For the cases in which the two neural networks disagreed as to the presence of a HIF current, the output of the two networks was summed and a variable threshold was used to make the decision. A threshold of 1.0 corresponded to making the final decision based upon which network was more confident in its own decision.

For example, if the output of network 1 was 0.9867 and the output of network 2 was 0.0175, then the sum would be less than 1.0 and a no HIF decision would be made because the output of network 2 is closer to the ideal value of 0 than the output of network 1 is to the ideal value of 1.

On the other hand, if a more conservative approach were desired in which one chose to reduce the false alarm rate, a larger threshold approaching 1.5 could be selected. In essence, a larger threshold gives more weight to the network that indicates a no HIF situation.

The network using the spectrum (FFT) of the monitored current would appear to be more capable of detecting HIF than the network using the actual current samples. Using the sampled current network in tandem with the spectrum based network can reduce the false alarm rates, however, it doesn't appear to increase the detection rate significantly. The lack of synchronizing the current's zero-crossing during training and generalization may prohibit this neural network from detecting some of the patterns or features attributed to HIFs, such as asymmetry of half cycles and variations from cycle to cycle.

Referring once again to Figure 2 there is shown an exemplary composite HIF detection system 12 that includes all of the three different techniques described above.

As can be seen from the test results of the three different HIF techniques 22, 24, 26 described above, none of them can detect all HIF faults while assuring no false alarms. The present invention evaluates the presence of HIF fault with all the above techniques and uses a multi-resolution framework having a decision logic 32 to detect the presence of high impedance fault. A fault is identified as a high impedance fault once it is independently detected by any two of a plurality of individual high impedance fault detection systems. An exemplary decision logic is described below:

if (Technique 1 = true); and if ((Technique 2 = true) OR (Technique 3 = true)), then HIF = TRUE end; else, if (Technique 1 = false); and if ((Technique 2 = true) AND (Technique 3 = true)) then HIF = TRUE end end. where, Technique 1 is the logical output (true or false) from the wavelet based algorithm; Technique 2 is the logical output from the algorithm based on higher order statistics; and Technique 3 is the logical output from the ANN based technique. For the above example, the logical output of any individual technique is true if that technique detects an HIF, otherwise it is false.

In order to reduce to ensure a good HIF detection performance by making the threshold calculation in each of the algorithms independent of the load a dynamic energy threshold calculation can be used according to the solution described in US patent 7,085,659 serial code 10/966,432 filed on October 15, 2004 whose content is hereby incorporated by reference. As shown for example in figure 1 1, an input signal comprising of phase (load) currents and/or neutral (residual) current, is input to the HIF detection algorithm 18 for processing. The HIF detection algorithm 18 may be one of the three algorithms previously described. The output of the HIF algorithm 18 is the energy of the input signal. This input signal energy is then multiplied by a factor, called Threshold Margin 14, that can be set to anywhere from about 1 10% to about 300% depending on the security of detection required and the result of that multiplication, known as Threshold Energy, is stored into a First-In First-Out (FIFO) buffer and control logic 13. Trials with captured field data indicate that there may be an unacceptable number of false detections when using a Threshold Margin lower than 125%. In general, the user of this technique would increase the Threshold Margin if the protected line has normally large and quickly varying frequency components of interest and the user wanted to reduce the probability of false detection. The FIFO buffer 13 has N elements and each element is updated every T seconds. The total delay from the input to the output of the buffer 13 is T*N seconds. The values used for T and N in one embodiment of the present invention are T = 10 seconds and N = 8 for a total delay through buffer 16 of 80 seconds or one (1) minute and 20 seconds. The updating period, T, is in that one embodiment selected as 10 seconds because it is the shortest time that produced acceptable detections given the sampling rate of 32 samples per cycle (about 2 kHz) in that embodiment. The value of 8 for N in that one embodiment is chosen to give the desired separation in number of minutes, one (1) in that embodiment, of lead-time between the present calculated energy and the Threshold Energy value: N = (number of minutes * 6) + 2 [where six (6) is the number of 10 second intervals in one (1) minute]. This value for N provides a clear distinction between pre-fault and fault values. The number of minutes or unit of time should be the maximum amount of time that it is expected to detect the fault. After that time expires, the fault energy begins to appear in the Threshold Energy which then makes detection less and less likely. The number of minutes or unit of time should be short enough that the HIF algorithm 18 can track normal changes in the load. Any element of the FIFO buffer 13 can be used as the threshold energy and is compared at 15 to the present energy signal. In one embodiment the three oldest values of the FIFO buffer 13, that is the three oldest values of the Threshold Energy, are used in a filter (not shown) to produce the one threshold value. There is a tradeoff between keeping enough elder values to provide sufficient time for detection versus keeping even more elder values and not using them which wastes memory. The filter provides for a smoother transition of the threshold outputs and because the data is updated so slowly (once every 10 seconds), any type of low- pass filter should be adequate to perform that function. When the input signal energy has a value greater than the Threshold Energy, an HIF detection signal is generated and that signal can be used to raise an HIF detection flag by any means, not shown but well known to those of ordinary skill in the art. It should be appreciated that while the embodiment described above uses the three oldest values of Threshold Energy stored in buffer 13 as the input to the filter to produce the one threshold value used for comparison that any or all of the values in the buffer 13 can be used for that purpose. In that one embodiment it was decided to use a filter that was easy to implement and that filter happens to use only the three oldest values.

During a reset of the algorithm, as would occur during initialization, all elements in the FIFO buffer 13 are assigned a reset value. The reset value is a relatively large value that prevents the comparator 15 from being activated and thus prevents a false detection while the system adapts to the input signal it is monitoring. Since the largest Threshold Margin is 300% or three times the typical load value a suitable reset value might be 10 times the typical load value that is obtained from the field data.

During normal operation, a HIF detection signal is generated when the computed input signal energy is larger than the Threshold Energy. This detection signal causes all elements of the FIFO buffer 13 to be set to the present output Threshold Energy threshold value. This provides a type of seal-in for the detection since an algorithm that has picked up, that is detected a HIF, will not drop out because the next Threshold Energy in the FIFO buffer 13 is larger. This action also clears the threshold pipeline of any values that may have been influenced by the fault before the Threshold Energy was exceeded.

Claims

ClaimsWhat is claims is:

1. A method for detecting a high impedance fault in an electrical power line after the opening and closing of a breaker, the method comprising: providing a plurality of high impedance fault detection means; providing a decision means; using said plurality of high impedance fault detection means to make a plurality of independent determinations from a signal taken from said power line, said plurality of high impedance fault detection means employing a first threshold value for a first predetermined period of time after the close of the breaker and a second threshold value after said first predetermined period of time; generating outputs representative of said independent determinations; and using said decision means to determine whether said high impedance fault has occurred.

2. The method of claim 1 wherein each of said plurality of high impedance fault detection means make said independent determination from a signal taken from said power line.

3. The method of claim 1 wherein said first threshold value comprises one of a first stored value or a second stored value.

4. The method of claim 3 wherein said first stored value comprises a fixed value and said second stored value corresponds to a maximum measured signal of the power line.

5. The method of claim 4 further comprising using said first stored value if the time after the breaker opened is greater than a second predetermined period of time measured from the opening of the breaker and using said second stored value if the time after the breaker opened is less than said second predetermined period of time.

6. The method of claim 4 wherein said second threshold value is dynamically determined.

7. The method of claim 6 wherein said second threshold value is stored in a first-in first-out buffer.

8. The method of claim 1 wherein said decision means determines that said high impedance fault has occurred if any two or more of said outputs indicate that said high impedance fault has occurred.

9. A fault detection system connected to an electrical power line for detecting a high impedance fault after the opening and closing of a breaker, said high impedance fault detection system comprising: a plurality of high impedance fault detection systems operable to respectively make a plurality of independent determinations from a signal taken from the electrical power line, said plurality of high impedance fault detection systems employing a first threshold value for a first predetermined period of time after the close of the breaker and a second threshold value after said first predetermined period of time, to respectively generate outputs representative of said independent determinations; and decision means connected to said high impedance fault detection systems for determining whether said high impedance fault has occurred.

10. The system of claim 9 wherein said decision means determines that said high impedance fault has occurred if any two or more of said outputs indicate that said high impedance fault has occurred.

1 1. The system of claim 9 wherein said first threshold value comprises one of a first stored value or a second stored value.

12. The system of claim 1 1 wherein said first stored value comprises a fixed value and said second stored value corresponds to a maximum measured signal of the power line.

13. The system of claim 12 wherein said first stored value is used if the present time is greater than a second predetermined period of time measured from the opening of the breaker and said second stored value is used if the present time is less than said second predetermined period of time.

14. The system of claim 9 wherein said second threshold value is dynamically determined.

15. The system of claim 14 wherein said second threshold value is stored in a first-in first- out buffer.

16. A fault detection system connected to an electrical power line for detecting a high impedance fault after the opening and closing of a breaker, said high impedance fault detection system comprising: a high impedance fault detection system operable to detect a high impedance fault from a signal taken from the electrical power line, said high impedance fault detection system employing a first threshold value for a first predetermined period of time and a second threshold value after said first predetermined period of time, and wherein said second threshold value is dynamically determined by sampling said signal from the electrical power line, and said first predetermined period of time is at least the minimum amount of time required to acquire sufficient samples to dynamically determine said second threshold value.

17. The system of claim 16 wherein said first threshold value comprises one of a first stored value or a second stored value.

18. The system of claim 17 wherein said first stored value comprises a fixed value and said second stored value corresponds to a maximum measured signal of the power line.

19. The system of claim 18 wherein said first stored value is used if the time after the breaker opened is greater than a second predetermined period of time measured from the opening of the breaker and said second stored value is used if the time after breaker opened is less than said second predetermined period of time.

20. The system of claim 16 wherein said second threshold value is stored in a first-in first- out buffer.