EP0939903A1

EP0939903A1 - Device and method for analysing faults on networks

Info

Publication number: EP0939903A1
Application number: EP97947070A
Authority: EP
Inventors: Hubert Podvin
Original assignee: Anpico
Current assignee: Anpico
Priority date: 1996-11-19
Filing date: 1997-11-17
Publication date: 1999-09-08
Also published as: FR2756051B1; FR2756051A1; AU5225098A; WO1998022828A1

Abstract

The invention concerns a method for the automatic analysis of faults on a network, characterised in that it comprises steps consisting in: breaking down the electric signals measured on the network, into electric loads each corresponding to a portion of time during which the measured signals are stable or quasi-stable, on the basis of a self-adapted quiescent value by statistical analysis of said signals, and detecting each faulty leading-in and each faulty output on each of the detected phenomena.

Description

DEVICE AND METHOD FOR ANALYZING DEFECTS ON NETWORKS

The present invention relates to the field of devices and methods for analyzing faults in electrical distribution networks. The invention applies in particular, but not exclusively, to MV (High Voltage A) networks having a nominal voltage value between 1 kV and 50kV.

The structure of such a network is illustrated diagrammatically in FIG. 1 appended. We can see in this figure 1 an MV distribution network 10 comprising a three-phase transformer 12 whose primaries are connected to the output of a HTB (High Voltage B) network with nominal voltage greater than or equal to 50kV and whose secondaries are connected to respective lines

21, 22, 23, of a busbar 20 of the HTA network. The lines 21, 22, 23 of the busbar are respectively connected to three-phase starting lines

30.1 to 30.n. These are themselves generally provided with ramifications, not illustrated in detail in FIG. 1 appended to simplify the illustration, but shown schematically for one of the phases of a departure. In this FIG. 1, the parameters Zn designate the impedance of each node of this phase seen from the start.

The source station comprising the transformer 12 is equipped with a shunt circuit breaker 14. This is adapted to shunt to the earth a phase detected as a fault in the case of a single-phase fault. Such grounding effectively contributes to the extinction of a possible arc. Furthermore, each line 30.1 to 30.n is provided with a cut-off device 32.

The companies in charge of electrical distribution have been interested for many years in the problem of detecting and locating MV works with electrical insulation faults. These are in fact the site of short-circuits, the elimination of which by protective mechanisms leads to short-term interruptions in the supply of energy to customers.

A method called "Network Auscultation" has been developed to detect and locate these faults using perturbographs installed in source stations and time-stamped fault detectors installed in the network.

However today, the exploitation of data from these devices is almost completely manual and requires a very tedious work volume for rigorous exploitation.

In an attempt to improve the detection and localization of such faults, the Department of Studies and Research of Electricity of France (DER) has developed an expert system (LAURE) intended to automatically process files and give probable location (s) short circuits on monitored departures. However, according to the design of this expert system, all the events recorded by the pertubographers must be previously repatriated to an analysis site, most often via the PTT network, which induces significant connection costs. The average time interval between the appearance of the fault and its analysis is thus of the order of 48 hours. This automatic analysis method cannot therefore be used to help the operator to restart a feed if the short circuit has caused a final trip.

The main interest of this method is therefore only to allow the analysis of so-called "fugitive" events, that is to say eliminated by the automatic protections of the departure. Experience shows, in fact, that these events often precede a definitive onset. In this case, the location of the short-circuits highlights the weak points of the network and makes it possible to ensure their maintenance.

Thus it appears that the means known to specialists for detecting and locating faults on the power supply lines, are not entirely satisfactory.

The present invention now aims to improve the situation. This object is achieved according to the present invention by means of an automatic fault analysis method on an electrical distribution network, which comprises the steps consisting in:

- decompose the electrical signals measured on the network, into electrical regimes each corresponding to a portion of time during which the measured signals are stable or almost stable, on the basis of a self-adapted rest value by statistical analysis of said signals, and

- detect each faulty arrival and each faulty departure on each of the events detected. The present invention also provides a device for analyzing faults on an electrical distribution network, characterized by the fact that it comprises means for automatic analysis of faults located at the level of a source distribution station to allow operation in time. real electrical signals from the network, without requiring the transfer of these to a remote analysis station.

Other characteristics, objects and advantages of the invention will appear on reading the detailed description which follows, and with reference to the appended drawings, given by way of nonlimiting example and in which: - Figure 1 previously described illustrates schematically the structure of a power supply network,

FIG. 2 represents voltages on the three phases of a network, when a single-phase fault occurs,

FIG. 3 represents a histogram of these voltages for determining a resting voltage,

FIG. 4 represents the appearance of a current belonging to a feeder not affected by the fault,

- Figure 5 represents transitions on this current belonging to a departure not concerned by the fault, - Figure 6 represents the normalized transitions on such a current belonging to a departure not concerned by the fault,

FIG. 7 shows the appearance of a current belonging to a departure affected by the fault,

FIG. 8 represents the transitions on this current belonging to a start affected by the fault,

FIG. 9 represents the normalized transitions on such a current belonging to a start affected by a fault, FIG. 10 represents the shape of the currents when a circuit breaker closes,

- Figure 11 shows the values of the resulting normalized continuous component, and - Figures 12 and 13 schematically represent the flow diagram of the method of the invention.

In the context of this patent application, the following terms or expressions will have the following meanings:

. Incident: Event disrupting the proper functioning of a network.

. Reclosing cycle: Set of automatic actions carried out by a protection device to try to eliminate the disturbance detected.

. Self-extinguishing faults: Events lasting less than 100 ms which disappear before the protection functions.

. Fugitive faults: Events which require the functioning of the protections and are eliminated by the shunt circuit breaker or by an opening of the outgoing circuit breaker of approximately 300ms ("fast").

. Semi-permanent faults: Events which require the functioning of the protections and disappear during the 1st and 2nd slow reset.

. Permanent faults: Events which are not eliminated by the protection automations and require operator intervention. . Single-phase fault: Short circuit between one of the phases and earth.

Such a fault is generally more difficult to locate than a polyphase fault because the impedance of the faulty loop depends on the zero sequence component of the cable which is difficult to model.

. Polyphase fault: Short circuit between two or three phases either isolated or with earth.

. Scalable fault: Single-phase fault evolving in the same place in polyphase fault. . Double fault: Earth faults resulting from the evolution of a first single-phase fault so that the affected phase is not that concerned by the first fault and the second fault is not located in the same place as the first . The device according to the present invention comprises, on the one hand means adapted to detect the appearance of faults on electrical distribution networks and, on the other hand means adapted to locate these faults on the network.

It preferably uses the signals coming from sensors associated or integrated with a disturbance recorder, designed to measure the three phase voltages and the neutral current of the arrivals and the three phase currents and the zero sequence current (or even only the zero sequence current) for the departures .

The system according to the present invention is based on the following general design defined by the Applicant.

The analysis of events likely to occur on a power supply line can be reduced to a problem of situation recognition. However, the rules for associating a given group of measures with a particular situation suffer from many exceptions. As a result, the system of the invention exploits heuristics, that is to say rules which are verified in certain cases and which are completely ineffective in others. Linear processing, the strategy of which is irrevocable, would be ineffective since it would be unable to question its choices if a failure was noted downstream. The system of the invention is therefore based on an Expert System type approach.

In the context of the invention, absolute priority is given to reliability.

The Applicant indeed considers that it is better to give a result with absolute certainty in 70% of the cases rather than a less reliable result in 90% of the cases.

Consequently, when the device according to the present invention cannot lead to an identification and a localization of a fault, it transmits to the operator a "failure" message inviting the operator to use other investigative methods if necessary to locate a possible defect.

The system according to the invention exploits a set of heuristics modeling the rules of the problem treated. Each of them gives a diagnostic output and possibly modifies one or more data of the common set.

The diagnosis delivered by the device is either:

. success with possibility of several solutions

. failure. undetermined.

In the case of success, each of the solutions is analyzed by the device and its consequences explored until the reasoning leads to a final diagnosis or failure.

In the event of a failure, the device "goes back" to the last choice made (lower level having one or more solutions) and continues with a new choice or leads to a definitive failure if no choice is possible.

In the case of an "indeterminate" diagnosis, the device tries another heuristic relating to the same level of analysis or otherwise ends with a definitive failure.

We will first describe the means according to the present invention for analyzing the events detected on the power supply lines.

The method for automatic analysis of faults on power supply lines, in accordance with the present invention mainly comprises the steps which consist in:

- carry out a statistical determination of the departures affected by faults,

- carry out a determination of the feeders concerned by the analysis of the phase between the neutral current and the zero sequence current,

- operate the same analysis for arrivals, - cutting out the measured voltages and currents, in stable or almost stable regimes, by using rest values defined by statistical analysis,

- all with elimination of instabilities, in particular transient circuit-breaker transients and possibility of modifying the breakdown into operating modes in the light of the results obtained, if necessary.

First, the means according to the present invention have the particularity of requiring no threshold for analysis. Only the information contained in the event is used. The "rest" voltage, ie the network voltage preceding the event is statistically determined. Changes in the nominal voltage due to the regulation used (15600V for a 15kV network for example) are therefore automatically taken into account.

As will be explained below, in the context of the invention, the incoming voltages are cut into electrical regimes.

For this it is necessary to know the quiescent voltage, that is to say the non-fault voltage of each phase. It is from this resting voltage that the device of the invention automatically determines the thresholds used to detect the transitions between the regimes. Different methods can be used to obtain this resting voltage.

Preferably within the framework of the invention, the device creates a histogram of the three phase voltages during the entire event. This histogram classifies all the values taken by the voltages during the event in increments, for example of 100V. The resting voltage used is the central value of the class with the highest frequency. It corresponds to the most commonly encountered voltage, which joins reality because the sum of the durations of the default regimes is always less than the sum of the durations of the non-default regimes (seen from the voltages).

There is thus illustrated in FIG. 2 appended, by way of nonlimiting example, the voltages on the three phases of a network, over a period of time during which a single-phase fault appears, and in FIG. 3 the histogram of these tensions. The central value of the class with the highest frequency, in FIG. 3, is 11500V, value retained as a representation of the quiescent voltage, for the analysis.

On this basis, the device according to the present invention cuts the electrical signals measured by a pertubograph, in electrical regimes, that is to say in stable or quasi-stable regimes.

The electrical regimes constituting an event are determined only by analyzing the voltages and currents measured by the disturbance recorder, and constituting the analyzed file. However, in the context of the invention, it is possible to use logic signals from the disturbance recorder if the other information is lacking. This determination must be made with care because its accuracy depends on the accuracy of the location. It is particularly difficult because certain constraints imposed are almost incompatible with each other. Indeed, it is necessary to filter the signals as little as possible if one wants to determine with precision the temporal position of each of the transitions between regimes and if one wants to be able to detect self-extinguishers of short duration. On the other hand, certain causes of short circuits have the effect of adding a fairly large background noise to the signals measured, which makes cutting difficult. The method used in the context of the invention consists in:

. split the event period by period using a non-critical threshold without arbitrary reduction of the information, and

. perform an intelligent smoothing of the result obtained thanks to rewriting rules which analyze the context and possibly correct the diagnosis made for each of the periods.

This last part of the processing constitutes a small expert system by itself and involves general notions of signal processing and / or electrotechnical knowledge (monitoring of the evolution of the phases between signals for example). The purpose of this smoothing is in particular to eliminate instabilities, such as for example the transients due to circuit breaker operations.

Secondly in the context of the invention, the device detects the arrival or arrivals in default and the departure or departures in default on each of the events by using the analog signals recorded. The analysis thus adapts to the operating conditions.

Again, different techniques can be used to carry out this detection.

Preferably, the device according to the present invention implements a method for the statistical determination of the departures concerned.

This method does not use any electrotechnical knowledge. It is based on the finding established by the Applicant that, during an event, the transitions in the amplitude variations of the currents of a departure are few in comparison with the small random variations due to the "background noise" of the steady state.

In order to be able to count these variations and thus to discriminate an event constituted by a defect, from background noise, the device according to the present invention operates a change of scale of the measured signals and normalizes their absolute value by representing them by a number between 0 and 1 (1 being the value assigned to the maximum amplitude). Then the device automatically counts the number of values other than 0. The device considers that a fault is detected when the number of normalized transition values is less than a threshold, for example less than 10.

To illustrate this analysis technique, we have illustrated:

- in FIG. 4, the shape of a current belonging to a feeder not affected by a fault, and which therefore only reveals background noise,

in FIG. 5, the transitions (or derivative) of this current belonging to this departure not affected by this fault, and

- in FIG. 6, these normalized transitions from 0 to 1, while we have illustrated: - in FIG. 7, the shape of a current belonging to a departure affected by a fault,

in FIG. 8, the transitions (or derivative) of this current belonging to a departure affected by the fault, and - in FIG. 9, these normalized transitions from 0 to 1.

The comparative examination of FIGS. 6 and 9 clearly shows that counting the number of normalized transition values makes it possible to discriminate the background noise of FIG. 6 which has a large number of transitions, from a defect illustrated in FIG. 9 which presents only two main standard transitions.

The device according to the present invention also implements a method for determining the feeders concerned by the analysis of the phase between the neutral current and the zero sequence current. This method uses electrotechnical knowledge. To do this, the device analyzes the phase between the neutral current of the transformer and the zero sequence current of each of the feeders. More specifically, in the context of the present invention, two different tests are preferably used depending on the number of departures to be studied. In the first, the device tests whether this phase belongs to an interval, for example the interval [5 °, 59 °]. If the value of the phase belongs to this interval, the start is considered to be affected by the fault. This test works correctly when the starting number is low but poses problems when this number becomes large. In the second, if the number of departures studied is greater than or equal to 3, the device performs the average of the non-zero phases of the various departures, then compares this average to each of them. It is the departure (s) whose phase deviates most from the average which are considered as the departure (s) in default. The device according to the present invention also implements a method of statistical determination of the arrivals concerned which is based on the same principle as that described above for departures and implements the voltages of the arrival.

The device according to the present invention can use different techniques to eliminate instabilities.

Preferably, it endeavors in particular to identify the transient circuit breaker opening and closing transients to eliminate them from processing and thus avoid the detection of false faults. When closing a circuit breaker after fast or slow reclosing, a transient regime of more or less long duration (up to 100 ms) appears on the voltages and is often difficult to distinguish from a three-phase fault regime because there is a simultaneous voltage dip on the 3 phases. However, the Applicant has demonstrated that when the phase currents of the start are measured, this transient phase can be highlighted because it is accompanied by a continuous component which is damped until it becomes zero.

To highlight this continuous component, we illustrated in figure 10 the shape of the currents at the time of the closing of a circuit breaker (end of fast at 900ms) and in figure 10 the values of the resulting normalized continuous component.

In addition, in the context of the invention, the device must sometimes operate a correlation between different events. In particular, the classic reclosing sequence (fast +

2 slow) usually lasts about 35s. However, known disturbers cannot record such a sequence at once. The analysis device according to the invention is therefore suitable for processing several files (up to 3) considering that it is a single event. The device according to the present invention can be used to detect numerous events such as operating operations, high voltage faults, monitoring of the neutral impedance, etc. It can also be adapted to the neutral regime compensated.

Diagrams of the analysis according to the invention have been illustrated in Figures 12 and 13 appended.

More specifically, FIG. 12 illustrates the steps for determining faulty arrivals and departures by statistical method as described above.

According to the particular and non-limiting flowchart illustrated in FIG. 12: the determination of faulty departures is carried out when at least one faulty arrival is determined and the detection of a fault on all arrivals is assimilated to a fault in the upstream network HTB. FIG. 13 diagrams the steps for determining the electrical regimes, and for analyzing the phases as indicated above.

These steps are preferably accompanied by a consistency test between the faults detected respectively on arrivals and departures.

Of course Figures 12 and 13 correspond only to a very schematic illustration of the invention which in practice uses non-linear processing.

We will now describe the means of locating an event, in accordance with the present invention.

The general aim of these means is to give the probable positions of a short circuit which is the source of an analyzed event, taking into account the network topology.

In the case of on-board localization in real time in accordance with the invention, the most urgent faults to be analyzed are those which have resulted in a definitive triggering of a departure. The present invention thus provides an aid to the conduct of the network by making it possible to indicate to the operator the cut-off members that he must actuate in order to restart the supply of energy as quickly as possible. In this context, reliability must be maximum and precision comes second.

However, the device according to the present invention also allows the localization of fugitive faults (preventive action) so that maintenance agents can be sent to the field to try to determine the cause. Precision is more important here, but the priority remains reliability.

In theory, the distance from the short-circuit point to the substation can be given by measuring the loop impedance of the fault and the line impedance per km, with a relationship of the type: distance in km = (Total reactance / Reactance per km) Unfortunately, the networks are generally not homogeneous, that is to say that they consist of sections whose reactances per km are different (portions of overhead line and underground) . It is therefore necessary to browse the "tree" representing the start and determine the sections by comparing the start and end reactances to the calculated reactance.

The location device according to the present invention is based on the following considerations. The calculation of the reactance of a faulty line portion in the case of a polyphase event does not pose a problem because it is independent of the topology of the departure concerned. On the other hand, the equation allowing the reactance of a faulty line to be calculated in the case of a single-phase fault is very difficult to solve if we do not rely on the line impedance values for each of the sections. In what follows, we will give the formulas, the methods used in the context of the invention and the validity constraints of these calculations.

In the calculation of the impedance of the faulty loop, the notion of temporal positioning of the measurement periods used is fundamental. In the case of a single-phase fault, the invention calls for a measurement during the non-fault regime preceding the fault regime and for a measurement during the fault regime. In the case of a polyphase fault, on the contrary only the fault regime is used according to the invention. Thus, the choice of the measures to be used is essential and determines the accuracy of the location.

In practice, the calculations are carried out according to the invention over all the periods or combinations of periods of the incriminated signals, thereby determining an interval of belonging of the reactance of the faulty line portion. The device according to the invention is however suitable for eliminating the transients of evolution from one regime to another (see determination of the calculation ranges explained below).

In what follows, we will have to manipulate approximate quantities but for which a validity interval will have been determined, that is to say an interval between a minimum value and a maximum value. We will write for example

Z = [R min, R max] + j [L min, L max] or

• [R min, R max] is the validity interval of the resistance

• [L min, L max] is the validity interval of the reactance. We will write

Z = [R, R] + J [L, L

We can redefine the 4 operations on the intervals

A = B + C = [B + Ç_, B + C]

A = B-C = [B-C.B-Ç]

A = B x C = [min (B x Ç_, B x Ç, B x C, B x C), max (B x Ç_, B x Ç_, B x C, B x C]

A = B-C = [min (B-Ç, B-Ç, B + C, B-C), max (B + Ç, B + Ç, B-C, B + C)]

on the condition that [ç, C do not contain 0.

The implementation of this type of arithmetic is simplified by the use of an object oriented language like C ++ where it is possible to redefine the operators on a class of objects which will be here a class of complexes defined over intervals.

The loop impedance of a single-phase fault is calculated as follows. The case of a single-phase fault is the most difficult to solve because it requires the implementation of the symmetrical components of the loop impedance. The expression for total impedance is:

Z ya ₊ + Z ₇ i ₄ + - Z ₇₀ + - i 3p = 6 ^χ (VM-ICHxZd) -

3x (LM-ICH) -IR

with

Zd is the direct impedance of the faulty line portion, • Zi is the inverse impedance of the faulty line portion,

• Zo is the zero sequence impedance of the faulty line portion,

• p the resistance of the fault considered to be purely resistive,

• VM the complex vector representing the fundamental of the voltage on the phase concerned during the fault regime,

• ICH the complex vector representing the fundamental of the current on the phase concerned during the previous non-fault regime,

• IM the complex vector representing the fundamental of the current on the phase concerned during the fault regime, • IR the complex vector representing the fundamental of the zero sequence current of the start (vector sum of the 3 phase currents) during the fault regime.

As Z _d = Z, and that moreover the device according to the present invention, is only interested in the imaginary part of the two members of equality, it is possible to give a shape that is easier to handle.

6VM 6ZdICH

Im (Zo) = Im () - Im () - Im (2Zd)

3 (LM - ICH) - LR 3 (IM - ICH) - IR

either by posing

6VM

ZA =

3 (Dvl - ICH) - LR

and

Im (Zo) = Im (ZA) - Im (ZBZd)

ZA and ZB are independent of the starting topology.

It will be noted that since the reverse impedance is always equal to the direct impedance, the invention considers, in everything which follows, only the direct impedance.

The loop impedance of a polyphase fault is calculated as follows.

In the case of a polyphase fault, the calculation of the loop impedance is simpler because it is expressed by the formula:

with

• VMa and VMb are the complex vectors representing the fundamentals of the voltages on the phases concerned during the fault regime,

• IMa and IMb are the complex vectors representing the fundamentals of the currents on the phases concerned during the fault regime.

We can write in this case

and

ZB = 0

For this type of event, only the direct impedance to the km of cables is necessary. In the case of a three-phase fault, the impedance between 2 phases is used. The fault must be balanced.

The aforementioned quantities ZA and ZB, both in the case of a single-phase fault and in the case of a polyphase fault, depend only on the measurements made on the analyzed event. The device according to the present invention takes account of variations during the event while eliminating the transients which can cause errors. To do this, it uses a sorting method on all measures or combinations of measures:

1. It loads all the possible values of ZA and ZB into an array, 2. it calculates the average value of ZA (ZAmoy),

3. it searches for the value of ZA (ZAmax) whose imaginary part is the most distant from the imaginary part of the mean,

4. it calculates

Ex = | Re (ZA max) - Re (ZAmoy) | Ey = | lm (ZA max) - Im (ZAmoy) | ,

_. , Ex x 100,, _L Ey x lOO,,., ....

5. as long as> rancid sheet and -> rancid sheet, it eliminates

Re (ZAmoy) Im (ZAmoy)

ZA max of the list and we return in 2, 6. If no more measurements remain, it invalidates the location. The approximation of the values of direct impedance and zero sequence reactance of the cables is carried out as follows.

Generally the topological files from the databases available today contain, for each line segment only

• a brief indication of how the cables are laid:

• underground

• aerial

• twisted, “the conductor section in mm ² ,

• the type of metal,

• an indication on the technology which is not systematically informed, therefore unusable, and

• the value of the linear resistance and reactance of the section. However, the determination of the values of the direct impedance of the cables depends on several criteria:

• cable structure (presence of a screen, type of screen, thickness)

• the arrangement of the phases (distance between conductors).

As for the values of the zero sequence reactances of the cables, they depend on the characteristics of the soil and on the possible presence of other connections, metallic masses, etc.

To best approach the impedances of the cables used, it was therefore decided in the context of the present invention, to establish an initial database with, in correspondence with each type of cable encountered, an approximation of the direct impedance values as well as homopolar reactance.

In the case of overhead cables, we most often deal with bare conductors, therefore without a screen.

For this type of cable, the device according to the present invention therefore uses the following formulas:

or

R is the apparent resistance of the conductor in Ω / km,

L is the apparent self inductance of the conductor in Ω / km and equal to

_{Λ /} _ _{Λ 1} , 2am. .05 + .21n () ω l0- ³ Ω / km 'with

• am average distance between the conductor axes in mm,

• d diameter of the conductive core in mm, and

• ω is the pulsation at 50Hz or 100π. The zero sequence reactance does not depend on the nature of the cable and is expressed by

Lo = 4π x 50 x ln (-) ωlO ^"4 Ω / km

\ 779dam ²

In the case of underground and twisted cables there is generally a screen and this must be taken into account.

The formula that gives the values of the direct and inverse impedances is

or

• R is the apparent resistance of the conductor in Ω / km, • Re is the resistance of the screen in Ω / km,

• M is the mutual inductance between the core and the screen in Ω / km and equal to

OR

• am is the average geometric distance between the conductor axes in mm, • dm is the average screen diameter in mm, • L is the apparent self inductance of the conductor in Ω / km calculated as above. The formula which gives the values of the zero sequence impedance is

with

Za = R + 3Rs + j4 x τr x / km

or

• Rs is the resistance of the ground in alternating current: π ² ï 10 ^" Ω / km, and

• h is the equivalent depth of return to the ground,

659 - 10 ³ with Qs the ground resistivity which is approximated to 100Ω.m

for normal soil in a temperate zone. The zero sequence reactance is the imaginary part of the result:

However when one tries to apply these formulas one can run up against a major problem: the ignorance of certain parameters necessary to computation and which it is necessary to estimate in particular:

• the cable diameter of which only certain values corresponding to standard sections are known (50mm ² , 95mm ² , 150mm ² and

240mm ² . In this case, we can extrapolate these values using a conventional interpolation method and determine the theoretical value of the cable diameter for each of the cables used, • for overhead cables

• the distance between the conductors' souls. In this case we can consider that this distance can be between 500 and 3000mm, • for underground cables

• the resistance of the screen linked to its nature (Aluminum or Copper), its thickness (1 or 2 microns) and its diameter. In this case we can consider that this diameter is equal to the average diameter of the cable,

• the average distance between the conductors' cores. In this case we can consider that the cables are laid in a trefoil. This distance is then equal to the diameter of the conductor. The device according to the present invention is suitable for estimating the validity interval of the direct impedance and of the zero sequence reactance by calculating all the possible values by varying each of the hypotheses. The distance between the conductor cores is between 500 and 3000mm and the functions involving this quantity are not linear. For this the device divides this interval into 5 parts within which the function is linear. Then, for example, it performs all the calculations for the following parameters: • type of screen: Aluminum or Copper,

• screen thickness: 1 or 2 micron,

• distance between the conductor cores: 500, 1000, 1500, 2000, 2500 and 3000mm.

Then the device defines the validity intervals of the direct impedance and the zero sequence reactance of the different cables by taking as minimum value the minimum value found and as maximum value the maximum value found.

The device according to the present invention uses a topology file. It is a file which describes the structure of the departure. It describes each line segment by giving the type of cable, its section, its location

(aerial or underground) and its length. These files can be supplied by each center managing an electrical distribution network and are, in principle, updated regularly. They are extracted for example from a cartographic database.

Since the topology file does not systematically contain the values of the direct impedance and the zero sequence reactance of the cables, the device in accordance with the present invention can use a database which lists each of the cables used by the topology files. This database gives, for each of them, the validity range of direct resistance, direct reactance and zero sequence reactance. For example in the case of an overhead copper cable of 19mm ² , with the previous assumptions, the database can include a line: A CU 0190 .947 0.947 0.205 0.462 1.544 2.058 1.095

1.095, according to which:

• direct resistance is represented by the interval [.947, .947] Ω / km • direct reactance is represented by the interval [.205, .462] Ω / km

• zero sequence reactance is represented by the interval [1.544, 2.058] Ω / km

• the zero sequence resistance, added for future extensions, is represented by the interval [1.095, 1.095] Ω / km. To calculate the validity intervals of the direct impedance and the zero sequence reactance of a section, the device proceeds as follows:

1. It determines the intervals of validity of these quantities for the upstream end,

2. It determines the intervals of these values of these quantities for the downstream end.

The validity interval of the result is given by [minimum value of the upstream end, maximum value of the downstream end].

To determine the section or sections affected by a single-phase fault, the device traverses the tree representing the topology of the departure (as illustrated in FIG. 1 for one of the departures) and searches for the sections whose validity intervals are such that the set of solutions to the equation Im (Zo) = Im (ZA) - Im (ZBZd)

is not empty (the validity range of the left limb has a common part with that of the right limb).

The validity interval of the direct impedance (Z _d ) and the validity range of the zero sequence reactance (lm (Z ₀ )) of the section are determined beforehand.

The location of a polyphase fault is carried out by traversing the tree in the same way as in the case of the single-phase fault. The device determines the section (s) affected by the fault by looking for those such as all of the solutions of the equation

VMa - VMb

Z, = Ma - Mb

The validity interval of the direct impedance (Zd) of the section is determined beforehand. Note however that in the case of a three-phase fault, the impedances measured between the phases taken two by two must be close

(the defect must be balanced) otherwise the location may become random.

In the process which has just been described, those skilled in the art will appreciate the particular importance of the following characteristics: the method for solving the loop impedance calculation equation in the case of a single-phase fault ,

- the use of intervals making it possible to automatically take into account the inaccuracies due to the "quality" of the measured signals but also to determine their feasibility, - the method of "sorting" the periods used for the calculations.

The approximations made with the hypotheses mentioned above make it possible to guarantee a correct location on the departure concerned. However, the areas delimited by the location are sometimes quite large and the accuracy of the result may be low. To eliminate this difficulty, it is proposed within the framework of the present invention, a learning of the events having been located by the user to allow to restrict these zones by improving the precision.

The simplest learning method is to learn the direct impedance and zero sequence reactance values for each of the cables used. However, the number of examples available is quite limited and does not guarantee that all types of cables from the initial database are affected. In addition, this method requires at least one event per start which can be difficult to obtain. Instead of being placed at the "impedance" level, the device according to the present invention is placed at the "calculation hypotheses" level, that is to say that it determines a set of hypotheses such that the known events are located "at best". Thus the adjustment is made on the conditions of calculation of the impedances and not on the impedances themselves. Even more precisely among the various existing learning methods, the Applicant has selected an optimization method based on a genetic algorithm. This method brings together all the necessary qualities:

• possibility of handling nonlinear relations. The relationships which exist between the precision of the localization and the parameters which one wants to adjust are complex and difficult to formalize, • simplicity of design and implementation.

The optimization method based on the genetic algorithm is an iterative method which uses an analogy with the process of natural selection and genetic evolution. The research carried out tends to optimize a “strength” function by using a “population” of candidates called “individuals”. In fact, each "individual" is a solution to the problem. If several variables are used, the value of each of them specific to the envisaged solution is coded. The individual is the set of all variables placed end to end. The initial population is made up of a certain number of individuals drawn at random. This number remains constant in subsequent generations. Each iteration, called “generation”, consists in making this population evolve towards a “daughter” population where only the most efficient “individuals”, taking into account the “force” function, are reproduced. To produce a new "population" the algorithm generates 3 steps:

1. evaluation of the “strength” of each “individual”,

2. “reproduction” of each individual in the daughter population in proportion to its strength, and 3. to avoid a possible blockage due to an overly homogeneous population, individuals from the new population are recombined, according to a given probability, using “ genetic operators "such as" crossing "and" mutation ". This corresponds to a search in different directions. The main advantages of genetic algorithms are:

• the possibility of working in parallel on different possible solutions,

• only an objective force function is necessary,

• the selection and combination stages are based on probabilistic rules and maintain a global search for possible solutions.

We will now proceed to a description of the genetic algorithm used according to a nonlimiting embodiment of the present invention.

An individual can consist of a 16-bit word coded as follows: • Bit 15: Copper screen (at 1)

• Bit 14: Aluminum screen (at 1)

• Bit 13: screen thickness 1 micron (at 1)

• Bit 12: screen thickness 2 micron (at 1)

• Bits 11 to 6: minimum distance between the conductor cores in mm (500 minimum)

• Bits 5 to 0: maximum distance between the conductor cores in mm (3000 maximum). This set of parameters is used to calculate the direct impedances and zero sequence reactances of the cables used in the examples. Each individual is a solution and makes it possible to establish a database with which the device according to the present invention assesses the ability to solve the equations generated using the measures of each of the available examples. Depending on the type of event (single-phase or multi-phase), the equation is

Im (Zo) = Im (ZA) - Im (ZBZd) for single-phase,

, _ _ VMa - VMb. . . 0 Zd = for polyphase.

Ma - Mb

In both cases the left limb is called "the calculated interval"

(value determined for a given section) and the right member is called

"The measured interval" (value from the measurement). 5 We arrive at interval comparisons. The performance of an individual will therefore determine how the solution he represents produces effective solution sets. The “force” function quantifies this efficiency.

This function is used to determine the strength of each individual, that is, the quality of the solution it generates. This force is given by the formula:

Force = Kl x Membership + K2 x Centering + K3 x Dispersion if Membership ≠ 0

Force = 0 otherwise where K1, K2, K3 are coefficients allowing to give more weight to one or the other of the elements constituting the result and which are: 5 • Membership: it measures the common part between the measured interval

X and the calculated interval Y. Its value will be higher the better the equations are solved. We define \\ X \\ as the "dimension" of the measured interval and | Y | the

"Dimension" of the calculated interval. It is the absolute difference between the minimum value 0 and the maximum value. With this definition we can write Intersection

Membership = x 100 min (| X ||, | Y ||) with Intersection defined as follows:

if max (X) <min (Y) or min (X)> max (Y) then Intersection = 0 otherwise Intersection = | max (min (X), min (Y)), min (max (X), max ( Y)) || Belonging tends to 100% when one of the two intervals is included in the other.

• Centering: it measures the distance between the center of the measured interval X and the center of the calculated interval Y (the center of an interval is the average of the minimum and maximum values). It is defined as follows: Centering = max (0, 100 - (d / max (| X |, || Y ||) x 100)

Centering tends to 100% when the centers of the two intervals are combined.

• Dispersion: it is the ratio between the dimension of the calculated interval Y and that of the measured interval X. It is calculated as follows: llxll - llYll if IIXll> jJYljalors Dimension = 100 - (" ^N ) x l00 || X ||

IIYII - llxll si | »X | I | l <I IIl Yl IIalors Dimension - 100 - (" _γ " j lOO

Dispersion tends to 100% when the dimension of the calculated interval tends to the dimension of the measured interval. It tends towards 0 when the dimension of one of the two intervals is much greater than the other.

The creation of a “daughter generation” from the current generation is carried out first by the step of calculating the strength of each of the individuals in accordance with the method described above. Then the device copies each of the individuals in the new generation in proportion to its strength with the following formula:

, Strength of the individual, _ΛΛ

Percentage of reproduction = x 100

Total force On the new generation it applies, with a probability fixed in advance

(10% for example), the crossing operator. On the result it applies, with a probability fixed in advance (10% for example), the mutation operator.

It then obtains the “daughter” generation which can itself give birth to a new generation, etc. The “crossing” operation is a mixture between two individuals with respect to a bit whose rank is chosen randomly between 0 and 15

Example: 2 individuals X ₁₅ , ..., X ₀ and Y15, ..., Yo. The draw designates the bit

5. After crossing the two individuals will be X15, ..., Xε, Y5, • -, Yo and Y ₁₅ If the crossover involves the same individual, it has no effect.

The mutation operation is the inversion of a bit whose rank is drawn randomly between 0 and 15

Example: 1 individual: 1010000000111111. Row 3 is drawn. By mutation, the individual 1010000000110111 will be obtained. It will be noted that the crossing and the mutation can generate individuals who are not "valid". In this case the procedure is canceled and we start again.

The genetic algorithm presented above having generated a set of more “tightened” hypotheses, we can then create a database

"Learned" with the formulas for calculating the previous impedances. In the case of learning examples, there is always a risk of

"Memorizing", that is to say that the system recognizes very well the examples that it has been taught but has difficulty recognizing a new example. If there is an improvement in the accuracy of the location, there is a risk of deterioration of the reliability. To avoid this degradation, the device according to the present invention retains the location on the initial database and superimposes the location on the "learned" database.

In the event that several sections of the network are identified as being likely to be affected by a fault according to the aforementioned technique, an ambiguity can be removed by using fault detectors distributed over the network in order to retain only the section actually concerned. Thus for example with reference to FIG. 1, in the hypothesis where two sections situated respectively between the nodes Z12-Z14 and Z22-Z24 would be identified as being liable to be affected by a fault, the analysis of the signals from sensors 40, 42 placed on the main starting trunk make it possible to retain only one of these two sections insofar as the sensor 42 "sees" the fault only if the sections Z22-Z24 are concerned.

In conclusion, the device according to the present invention allows real-time processing of events, as they occur and the immediate sending of a message to the operator. The device of the invention thus makes it possible to combine a curative treatment with a preventive treatment. Thanks to the means of the invention, the indication to the operator of the probable areas of a short circuit on the feeder may not exceed 5 mm from the appearance of the fault and makes it possible to save considerable time for return to service.

Of course the present invention is not limited to the particular embodiment which has just been described, but extends to any variant in accordance with its spirit.

Claims

1. Method for automatic analysis of faults on an electrical distribution network, characterized in that it comprises the steps consisting of:

- decompose the electrical signals measured on the network, into electrical regimes each corresponding to a portion of time during which the measured signals are stable or quasi-stable, on the basis of a self-adapted rest value by statistical analysis of said signals, And

- detection of each faulty arrival and each faulty departure on each of the detected events.

2. Method according to claim 1, characterized in that the rest value is obtained by creating a histogram of the measured voltages, in predetermined increments, for example of 100V, the central value of the class whose frequency is the highest. high being retained as the resting value.

3. Method according to one of claims 1 or 2, characterized in that the signal decomposition step consists of: - dividing each event period by period in relation to a non-critical threshold, and

- carry out a smoothing of the result obtained using rewriting rules to possibly correct the division made over each of the periods.

4. Method according to one of claims 1 to 3, characterized in that the detection sequence comprises a step of statistical determination of the departures affected by a fault.

5. Method according to one of claims 1 to 4, characterized in that the detection sequence comprises a step of statistical determination of arrivals affected by a fault.

6. Method according to one of claims 4 or 5, characterized in that the statistical determination step consists of normalizing the transitions of amplitude variations of the signal and counting the number of standardized transitions obtained, a fault being considered detected when the number of normalized transition values is less than a threshold, for example less than 10.

7. Method according to one of claims 1 to 6, characterized in that it comprises a step of analyzing the phase between the neutral current and the zero sequence current.

8. Method according to claim 7, characterized in that the step of analyzing the phase between the neutral current and the zero sequence current consists of: - for a small number of departures studied, for example less than

3, to test if the phase belongs to a determined interval, for example from 5° to 59°, in which case the channel tested is considered to be affected by the fault, and

- for a large number of departures studied, for example greater than or equal to 3, to take the average of the non-zero phases, compare this average to each phase and consider as default the departure(s) whose phase is furthest from the average.

9. Method according to one of claims 1 to 8, characterized in that it comprises a step consisting of eliminating the instabilities detected on the signal.

10. Method according to claim 9, characterized in that the step of eliminating instabilities consists of eliminating transients due to circuit breaker operations.

11. Method according to one of claims 8 or 9, characterized in that the step of eliminating instabilities consists of detecting the periods during which a continuous component appears and eliminating these periods from the location calculation.

12. Method according to one of claims 1 to 11, characterized in that it comprises a correlation step between different events, such as for example the successive events of a reactivation sequence.

13. Method according to one of claims 1 to 12, characterized in that it further comprises a step of locating a detected fault.

14. Method according to claim 13, characterized in that for the detection of a single-phase fault, the location step consists of traversing the topology of a departure by searching for sections whose validity intervals are such that the set of solutions of the equation

Im(Zo) = Im(ZA) - Im(ZBZd)

is not empty, that is to say that the validity interval of the left side has a common part with that of the right side, a relationship in which:

6VM

ZA =

3(M - ICH) - LR

And

Zd is the direct impedance of the faulty line portion,

Zo is the zero sequence impedance of the faulty line portion,

VM the complex vector representing the fundamental of the voltage on the phase concerned during the fault regime, ICH the complex vector representing the fundamental of the current on the phase concerned during the previous non-fault regime,

IM the complex vector representing the fundamental of the current on the phase concerned during the fault regime, and

IR the complex vector representing the fundamental of the starting zero sequence current (vector sum of the 3 phase currents) during the fault regime.

15. Method according to one of claims 13 or 14, characterized in that for the detection of a polyphase fault, the location step consists of traversing the topology of a departure by searching for the sections whose validity intervals are such that the set of solutions of the equation

is not empty, that is to say the validity interval of the left side has a common part with that of the right side, a relationship in which:

• VMa and VMb are the complex vectors representing the fundamental voltages on the phases concerned during the fault regime,

• IMa and IMb are the complex vectors representing the fundamental currents on the phases concerned during the fault regime.

16. Method according to one of claims 1 to 15, characterized in that to take into account variations during the event while eliminating transients which can cause errors, it implements a sorting process on all the measurements or combinations of measures.

17. Method according to claim 16, characterized in that the sorting process consists of:

1. load all possible values of ZA and ZB into a table, 2. calculate the average value of ZA (ZAmoy),

3. find the value of ZA (ZAmax) whose imaginary part is furthest from the imaginary part of the average,

4. calculate

Ex = |Re(ZA max) - Re(ZAmoy)| Ey = |lm(ZA max) - Im(ZAmoy)| ,

_ . . Ex x 100, , _t Ey x lOO, , ... .

5. as long as > rancid sheet metal and — > rancid sheet metal, eliminate

Re(ZA oy) Im(ZAmoy)

ZA max of the list and return to 2, and

6. If no more measurements remain, invalidate the location.

18. Method according to one of claims 1 to 17, characterized in that it further comprises the step consisting of establishing an initial database with, in correspondence with each type of cable encountered, a approximation of direct impedance values as well as zero sequence reactance.

19. Method according to claim 18, characterized in that for aerial cables, the database is established using the following formulas: for the direct impedance value:

or • R is the apparent resistance of the conductor in Ω/km,

• L is the apparent self-inductance of the conductor in Ω/km and equal for example to

... . ,2am.

L = .05+.2ln( ) co lO ^~3 Ω / km ' with

• am average distance between the conductor axes in mm,

• d diameter of the conductive core in mm, and

• ω is the pulsation at 50Hz or 100π, and

• for zero sequence reactance:

2h ³

Lo = 4π x 50 x ln(- )ωlO ^~ Ω/ km

779dam-

20. Method according to one of claims 18 or 19, characterized in that for underground or twisted cables, the database is established using the following formulas: for the direct impedance:

Or • R is the apparent resistance of the conductor in Ω/km,

• Re is the screen resistance in Ω/km,

• M is the mutual inductance between the core and the screen in Ω/km and equal for example to

2a™ M ≈.21n ω lO ^_3 Ω / km

Or

• am is the average geometric distance between the conductor axes in mm,

• dm is the average diameter of the screen in mm,

• L is the apparent self-inductance of the conductor in Ω/km calculated as previously, and for the zero-sequence impedance:

~ J a Ω/km

with

Za m

Or

• Rs is the ground resistance in alternating current, and

• h is the equivalent depth of return in the ground, i.e. .

• for normal soil in a temperate zone.

21. Method according to one of claims 1 to 20, characterized in that it uses a topology file which describes the structure of the departure, for example by giving the type of cable, its section, its location (aerial or underground ) and its length.

22. Method according to one of claims 1 to 21, characterized in that to calculate the validity intervals of the direct impedance and the zero sequence reactance of a section it proceeds as follows: i) by determining the intervals of validity of these quantities for the upstream end, ii) by determining the intervals of these values of these quantities for the downstream end, and iii) by retaining the validity interval given by [minimum value of the upstream end, value maximum of the downstream end].

23. Method according to one of claims 1 to 22, characterized in that it implements a process of learning localized events to improve the precision of detection.

24. Method according to claim 23, characterized in that the learning process operates by adjusting the conditions for calculating the impedances, by determining a set of hypotheses such that the known events are located optimally.

25. Method according to one of claims 23 to 24, characterized in that the learning process implements a genetic algorithm.

26. Method according to one of claims 1 to 25, characterized in that it implements a double location: based on an initial database comprising an approximation of the direct impedance values and the zero sequence reactance of each section, on the one hand, and based on a database resulting from learning, on the other hand.

27. Method according to one of claims 13 to 26, characterized in that it further comprises a step of removing ambiguity on the location of a fault, by exploiting signals from sensors distributed over the network.

28. Device for analyzing faults on an electrical distribution network, characterized in that it comprises means for automatic analysis of faults located at a distribution source station to allow real-time exploitation of electrical signals from the network, without requiring their transfer to a remote analysis station.

29. Device according to claim 28, characterized in that it comprises on the one hand means adapted to detect the appearance of faults on an electrical distribution network and on the other hand means adapted to locate these faults on the network.

30. Device according to one of claims 28 or 29, characterized in that it uses the signals coming from sensors associated with or integrated into a disturbance recorder, designed to measure the 3 phase voltages and the neutral current of the arrivals and the three phase currents and the zero sequence current (see only zero sequence current) for the feeders.

31. Device according to one of claims 28 to 30, characterized in that it comprises: - means capable of breaking down the electrical signals measured on the network, into electrical regimes each corresponding to a portion of time during which the signals measured are stable or quasi-stable, on the basis of a self-adapted rest value by statistical analysis of said signals, and - means for detecting each faulty arrival and each faulty departure on each of the detected events.

32. Device according to claim 31, characterized in that the means adapted to define the rest value are adapted to create a histogram of the measured voltages, in predetermined increments, for example of 100V, and retain the central value of the class of which the frequency is highest as a resting value.

33. Device according to one of claims 31 or 32, characterized in that it comprises:

- means capable of dividing each event period by period in relation to a non-critical threshold, and

- means capable of carrying out a smoothing of the result obtained thanks to rewriting rules to possibly correct the division carried out over each of the periods.

34. Device according to one of claims 28 to 33, characterized in that it comprises means capable of carrying out a statistical determination of the departures or arrivals affected by a fault, which means are adapted to normalize the transitions of variations of amplitude of the signal and count the number of normalized transitions obtained, a fault being considered detected when the number of normalized transition values is less than a threshold, for example less than 10.

35. Device according to one of claims 28 to 34, characterized in that it comprises means for analyzing the phase between the neutral current and the zero sequence current.

36. Device according to one of claims 28 to 35, characterized in that the means for analyzing the phase between the neutral current and the zero sequence current are adapted: - for a small number of departures studied, for example lower has

3 to test whether the phase belongs to a determined interval, for example from 5° to 59°, in which case the channel tested is considered to be affected by the fault, and

- for an elaborate number of departures studied, for example greater than or equal to 3, to take the average of the non-zero phases, compare this average to each phase and consider as a default the phase which is furthest from the average.

37. Device according to one of claims 28 to 36, characterized in that it comprises means capable of eliminating the instabilities detected on the signal, in particular the transients due to circuit breaker operations.

38. Device according to claim 37, characterized in that the means for eliminating instabilities are adapted to detect the periods during which a continuous component appears and eliminate these periods from the location calculation.

39. Device according to claim 38, characterized in that the location means are adapted to browse the topology of a departure by looking for the sections whose validity intervals are such that the set of solutions of the equation

Im(Zo) = Im(ZA) - Im(ZBZd)

6VM

Z.A.

3(M - ICH) - LR

And

Zd is the direct impedance of the faulty line portion, Zo is the zero sequence impedance of the faulty line portion,

VM the complex vector representing the fundamental of the voltage on the phase concerned during the fault regime,

ICH the complex vector representing the fundamental of the current on the phase concerned during the previous non-fault regime, IM the complex vector representing the fundamental of the current on the phase concerned during the fault regime, and

40. Device according to one of claims 38 or 39, characterized in that the location means are adapted to browse the topology of a departure by searching for sections whose validity intervals are such that all of the solutions of the equation Im(Zo) = Im(ZA) - Im(ZBZd)

• VMa and VMb are the complex vectors representing the fundamentals of the voltages on the phases concerned during the fault regime, • IMa and IMb are the complex vectors representing the fundamentals of the currents on the phases concerned during the fault regime.

41. Device according to one of claims 28 to 40, characterized in that it comprises means capable of establishing an initial database containing, in correspondence with each type of cable encountered, an approximation of the direct impedance values as well as zero sequence reactance.

42. Device according to one of claims 28 to 41, characterized in that it uses a topology file which describes the structure of the departure, for example by giving the type of cable, its section, its location (aerial or underground ) and its length.