Sie sind auf Seite 1von 4

EMC’14/Tokyo 16P2-B5

Multilayer Ground Determination from Apparent


Resistivities and Impact on Grounding Resistances
Gabriel Papaiz-Garbini1,2, Lionel Pichon2, Michel Cucchiaro1 and Noel Haddad1
1
SNCF Engineering, Electromagnetic Compatibility Service, La Plaine St Denis, France
2
Laboratoire de Génie Électrique de Paris, UMR 8507 CNRS, SUPELEC, Université Paris-Sud, Université Pierre et Marie Curie,
Gif sur Yvette, France
gabriel.papaiz-garbini@sncf.fr, lionel.pichon@lgep.supelec.fr, michel.cucchiaro@sncf.fr, noel.haddad@sncf.fr

Abstract—The determination of the ground structure in terms This paper will firstly present the importance of taking into
of its electric parameters (resistivity and permittivity) is a key account a multilayer ground by presenting an example of
point for calculations of ground electrical conduction, especially grounding calculation using the finite element software
when high voltages are involved. A mixed optimization method COMSOL®.
designed to determinate the resistivities and lengths of the
ground layers in a horizontally stratified model for the soil is Next, a simulation of a Wenner method measurement over
presented, using the Genetic Algorithms and Pattern Search. The a chosen soil, defined as in [9], was performed by using
input for this method will be a set of apparent resistivities, COMSOL®. This simulation has been designed to reproduce
measured using the four-probe Wenner method. This paper will the effects of a real soil. Indeed, the Wenner method is one of
justify the need of taking into account a heterogeneous soil. the most used methods to measure earth resistivities from
surface prospecting [10]. From a set of simulations from
Keywords—Wenner method; genetic algorithm; pattern different Wenner electrodes spacing, a set of apparent
search; multilayered ground; finite elements resistivities is obtained.
From the apparent resistivities values evolution with
I. INTRODUCTION respect to the Wenner electrodes spacing, the number of
One of the most dangerous consequences of a hazardous ground layers that would suit the best our measurements can be
short-circuit in a power transmission system is the emergence determined, following [11]. An inverse calculation enables us
of a Ground Potential Rising (GPR), as the short-circuit current to find the true electric parameters of the soil that would
is normally conducted to the soil. Usually undervalued, the originate the apparent resistivities calculated. In this aim, a
GPR arising from high voltage power system might produce combination of two optimization methods has been used: a
high step and touch voltages, putting in danger people nearby genetic algorithm and a pattern search methodology. The first
[1]. Moreover, when it occurs in the vicinity of the railway one is used to calculate a first good approximation of the result
system, the railway conductors participate actively in the that will be given as an input to the second one. This
diffusion of the current [2]. methodology has already been applied in a different scientific
domain, such as in [12]. Both optimization algorithms were
To prevent these dangerous voltages, some numerical used from the Optimization Toolbox of Matlab©.
methods were specially designed to calculate the GPR that
follows a short-circuit in a power system. For this kind of
II. THE NEED OF A MULTILAYER SOIL
calculation, the most difficult part of the problem is the electric
conduction through the ground itself, and this is mostly
A. Grounding calculation method
because of our lack of knowledge of what happens
underground [3]. For this reason, historically, the presence of A real four layer soil, for which electric parameters have
the soil has been simply neglected or oversimplified in many been measured in a real scenario by [9], has been chosen. For
cases and as much as possible. this type of soil, the apparent resistivity that one would
measure by doing a Wenner experiment, for different Wenner
One of the first important studies where the effects of the electrodes spacing, has been calculated using COMSOL®.
soil have been taken into account has been done by Carson [4],
where he treats the impact of the ground on the impedance of The principle of the Wenner method (Fig. 1) consists of
overhead conductors. The first work fully dedicated to the four probes that are buried and separated from each other by
subject of the conduction of the ground was published by the same distance a. An electrical current is injected and
Sunde [5]. Many models have been used since then for the collected from the two probes at the extremities, while the two
electric conduction in the soil, most of them considering the in the center measure the electrical potential difference.
ground as an homogeneous conductor, such as in [6], or as a From the current I and voltage V values, the apparent
two parallel layer ground, as in [7]. In order to have a better resistivity is found by supposing that we have homogeneous
description of the ground, the use of genetic algorithms as an soil and that each current electrode induce separately a
optimization method has been proposed by [8]. variation of the electric potential in each one the voltage
electrodes. We will have:

Copyright 2014 IEICE


789
EMC’14/Tokyo 16P2-B5

V radius and 2m of length. The top of the electrode is in the


ρ app = 2πa (1) same level as the top of the first ground layer.
I Taking into account the apparent resistivities calculated
before, the grounding resistance of this electrode has been
where a is the distance of two adjacent Wenner electrodes. calculated in the three different scenarios described before.
Those values are presented in TABLE II.

TABLE II. GROUNDING RESISTANCE CALCULATION

Scenario 1 Scenario 2 Scenario 3


Grounding Resistance (Ω) 682 2874 11227
Percentage Difference to
- 321 % 1546 %
Fig. 1. Four probes Wenner method. Scenario 1

To compare different scenarios, three different types of


ground with different resistivities have been defined: From this first result one can see that the grounding
resistance calculation is the basis for the safety systems
• Scenario 1: the complete 4-layer ground as in Fig. 2. implementation in terms of current diffusion in case of short-
circuits. A grounding circuit poorly designed can lead to
• Scenario 2: an homogeneous soil of resistivity being
dangerous step and touch voltage and also to equipment
the arithmetical mean of three different apparent
destruction. In other words, the grounding resistance being the
resistivities calculated with three different values of d.
capacity of interaction between the conductor and the ground,
• Scenario 3: homogeneous soil of resistivity being the this calculation shows that a reliable description of the electric
resistivity of the first layer. parameters of the soil is needed.
The grounding resistance value for a specific electrode is
defined and calculated by COMSOL as being the ratio between III. THE OPTIMIZATION METHODS
a chosen voltage, and the amount of electrical current that it In order to find the characteristics of a ground that could
can evacuate in the soil. give rise to a certain vector of apparent resistivities, two
different optimization methods have been used: Genetic
B. Results Algorithms and Pattern Search. It allowed us to increase
performance by giving a better precision and by accelerating
The modeled ground structure can be seen in Fig. 2:
convergence.
Although the genetic algorithm normally converges to a
point close to the solution, almost independently from the
starting point, from a certain degree of precision the
convergence can take too much time [13]. The pattern search is
used for increasing the reliability of the genetic algorithm, the
output of the latter being the input of the former.
These optimization methods require the use of a certain
function to be minimized, which is called objective or fitness
Fig. 2. Soil structure for a four parallel layer modelling. In this case, he have function. For our problem, the objective function has been
7 parameters to be determined.
borrowed from [14], who gives us an equation for the apparent
For three different Wenner electrode distances, the resistivities of a Wenner measurement over a N-layered soil
COMSOL® simulation of the Wenner method has been from the resistivities and sizes of all the layers.
performed. TABLE I. presents the apparent resistivities In the next sections, we will give a brief description of
calculated. these two algorithms.

TABLE I. APPARENT RESISTIVITY CALCULATIONS A. Genetic Algorithms


Wenner electrode spacing (m) Apparent Resistivity (Ω.m)
The genetic algorithms is an optimization method that
reproduce the principles of natural selection, meaning that the
2.0 2602 individuals with the best characteristics (which means, with a
8.0 397 smaller objective function associated to it) are those who have
the biggest probability to transmit their genes to the next
16.0 380 generation [13].
The first step of the algorithm is to randomly determine an
For the calculation of the grounding resistance, the ground initial population of individuals. Each individual is composed
electrode has been chosen as made of cooper with 50cm of of a vector of numbers which are those we want to optimize.

Copyright 2014 IEICE


790
EMC’14/Tokyo 16P2-B5

These numbers are called the “genes”. The population will thickness h1= 10m , while the second layer has a resistivity of
evolve by combining the genes of different individuals. In this
aim, three mainly methods are used: ρ 2 = 40Ω ⋅ m and an infinite thickness.
• Crossover: where two individuals randomly chosen
combine their genes to create a new one.
• Mutation: where some genes of one individual are
randomly changed.
• Elite child: where some randomly chosen individuals
are passed unchanged to the next generation of
individuals.
We have used the algorithm from the Toolbox of Matlab
[15]. The proportion of individuals of the next generation
produced by each of the evolving methods is defined by the
user. In our case, it has been chosen by trial and error, in order Fig. 3. Hybrid optimization calculation for a 2-layer ground.
to enhance performance, a rate of 40% of crossover, 5% of
elite individuals and 55% of mutations. The Fig. 3 shows the evolution of the three ground
parameters. The convergence is really fast, taking less than one
B. Pattern Search Method minute in a Core i7 Intel laptop. The first 200 iterations have
The pattern search methods are a group of direct been done by the genetic algorithms, which gives us a
optimization methods that, unlike more traditional determinist precision that is better than 1% of error.
optimization methods, does not require the calculation of any After the pattern search method, which represents the last
derivative of the objective function [16]. one hundred iterations, the error was reduced to almost zero.
From an initial point in the space of the possible solutions For all the parameters, it was, at most, less than 0.1%.
of the problem, the algorithm draws a regular pattern around
this point. Next, it calculates the value of the objective function B. Case 2: 3 Layer Soil
on all the vertices of the pattern. Among all these points, the For the 2 layer case, we have chosen to model a first layer
algorithm chooses the best one, which will be the initial point having a resistivity of ρ1 = 100Ω ⋅ m and a
for the next iteration.
If the algorithm does not find any point that is better than
thickness h1 = 10m , a second layer having a resistivity of
the initial one, it creates a new smaller pattern around this point ρ 2 = 40Ω ⋅ m and a thickness h2 = 20m , and a third layer
in order to search to another, closer to it. When the size of the
pattern is smaller than a certain limit, the algorithm stops. with of ρ 3 = 1000Ω ⋅ m and an infinite thickness.

For two first layers parameters, the convergence is as fast


IV. RESULTS as for the previous case, as shown in Fig. 4. The genetic
The optimization methods were applied in three different algorithm is used for the 1200 first iterations, while the pattern
situations. The first two cases are based on simulations of a search continues after that. The error is also less than 0.01%.
Wenner measurement of apparent resistivities of a chosen soil,
which has been modeled using COMSOL®. For the first one, a
2 layer soil has been chosen, while for the second one it has
been a 3 layer soil. In both configurations, the Wenner
electrode spacing comes from the following input vector:
[ ]
a = 2,4,6,8,10,12,14,16 . Moreover, as we know the exact
structure of the soil, we are able to perfectly determine the
percentage of error of the solutions presented.
The third case comes from a set of apparent resistivities
values that have been measured over an unknown soil in the
region of Nîmes, south of France. As the apparent resistivities
were taken directly from the measurements, no COMSOL® Fig. 4. Hybrid optimization calculation for a 3-layer ground, olnly showing
simulation was needed for this last case. the parameters of the two first layers.

A. Case 1: 2 Layer Soil For the third layer, as we can see in Fig. 5, the convergence
is slower, specially because of the small number of apparent
For the 2 layer case, we have chosen to model a first layer
resistivities that were used. Also, it is very important that the
having resistivity of ρ1 = 100Ω ⋅ m and distance of Wenner electrodes being larger than the size of the
layers that we want to find. So, to decrease simulation time and

Copyright 2014 IEICE


791
EMC’14/Tokyo 16P2-B5

increase precision, we should simulate more Wenner CONCLUSION


measurements and with larger electrodes spacing. The COMSOL® simulations demonstrated the need of a
Nevertheless, after some hours of calculations, it was more complete description of the soil in terms of its electric
possible to get a less than 5% error, as we can see in Fig. 5. parameters. Then, a new methodology, mixing genetic
algorithms and pattern search, has been proposed in order to
recover these parameters starting from a set of Wenner
apparent resistivities measurements. Simulations attested the
performance of the calculation.
The application of this methodology in real case scenarios,
particularly for the railway studies of electromagnetic
compatibility performed by SNCF, will increase the reliability
of the calculations.

REFERENCES
Fig. 5. Hybrid optimization calculation for a 3-layer ground, only showing [1] J.F. Laidig and F.P. Zupa "A practical ground potential rise prediction
the parameters of the two first layers. technique", IEEE Transactions on PAS, vol. PAS-99, pp.207 -216
1980
C. Case 3: Measurements in a real scenario [2] N. Haddad and M. Cucchiaro, "Ground Potentiel Rise Calculation
applied to a multiconductor network of a railway system", 2012 EMC
For the real case measurement, we do not have any Europe International Symposium, 2012
previous information about the ground composition. It prevents [3] D.L. Garrett and J.G. Pruitt: "Problems Encountered With The Average
us to actually quantify the error of the calculations. Potential Method of Analyzing Substation Grounding Systems", IEEE
Transactions on Power Apparatus and Systems, Vol. PAS-104, No. 12,
Nevertheless, it is possible to quantify the quality of the pp. 3585-3596, 1985
solution by comparing the measured apparent resistivities with [4] J. R. Carson, "Wave propagation in overhead wires with ground
a set of calculated apparent resistivities. The calculation of the return", Bell Syst. Tech. J., vol. 5, pp.539 -554 1926
apparent resistivities has been done following the same steps of [5] E. D. Sunde, Earth Conduction Effects in Transmission Systems, Dover,
[14], which has also been used for the calculation of the 1949
objective function. [6] S. J. Schwarz, Analytical Expression for the Resistance of Grounding
Systems, IEEE Power Apparatus and Systems, vol. 73, issue 2, part III-
The TABLE III. presents the comparison between the B, January 1954.
measured and calculated apparent resistivities for a three layer [7] J. Nahman, A pratical method for the interpretation of earth resistivity
soil simulation. The calculated electric parameters have data obtained from driven rod tests, vol.3, issue 4, pp. 1375-1379,
October 1988.
been ρ1 = 85.4Ω ⋅ m and h1 = 2.6m for the first layer, [8] W.P. Calixto, L.M. Neto, M. Wu, K. Yamanaka, E. da Paz Moreira,
Parameters Estimation of a Horizontal Multilayer Soil Using Genetic
ρ 2 = 1165Ω ⋅ m and h2 = 3m for the second layer, Algorithm, IEEE Transactions on Power Delivery, vol. 25, no. 3, pp.
1250-1257, 2010.
ρ 3 = 10 −4 Ω ⋅ m and an infinite thickness for the third layer.
[9] J-P. Nzuru Nsekere, Contribution à l'analyse et à la réalisation des mises
We notice the excellent agreement between the values. à la terre des installations électrique dans les régions tropicales, PhD
thesis of Université de Liège, February 2009.
[10] W. Ruan, R. D. Southey, S. Fortin, and F. P. Dawalibi, "Effective
TABLE III. MEASURED APPARENT RESISTIVITIES sounding depths for HVDC grounding electrode design: Wenner versus
Schlumberger methods, " in Dalian, China, IEEE/PES Transm. and
Electrodes Distrib. Conference and Exhihition, pp. 1-7, December 2005.
2 4 6 8 10 12 14 16
Spacing (m)
[11] Zohdy, A.A.R., A new method for the interpretation of Schlumberger
Measured
and Wenner sounding curves, Geophysics, 1989, 54:245–253.
Apparent 102.8 147.5 186.0 205.7 210.5 205.9 202.3 181.0
Resistivity (Ω.m) [12] J. L. Payne, M. J. Eppstein, A Hybrid Genetic Algorithm with Pattern
Search for Finding Heavy Atoms in Protein Crystals, Proceeding
Calculated
GECCO '05, 2005
Apparent 102.6 148.3 185.2 205.2 211.5 207.8 197.6 183.5
Resistivity (Ω.m) [13] J. Holland, Adaptation in Natural and Artificial Systems, The University
of Michigan Press, Ann Arbor, 1975.
Deviation (%) 0.2 0.6 0.4 0.2 0.4 0.9 2.3 1.4 [14] T. Takahashi, T. Kawase, Analysis of apparent resistivity in a multilayer
earth structure, IEEE Trans on Power Delivery, 1990,5(2): 604-612.
[15] A.J. Chipperfield and P.J. Fleming, The MATLAB genetic algorithm
However, in order to ensure that there are not multiple toolbox, Proceedings of the IEE Colloquium on Applied Control
solutions for the problem, one has to make Wenner Techniques using MATLAB, pp. 10/1-10/4, 1995.
measurements for larger distances. It comes from the fact that a [16] Abramson, Mark A. Pattern Search Filter Algorithms for Mixed
ground layer with a size L will only have an influence in the Variable General Constrained Optimization Problems. Ph.D. Thesis,
Wenner measurements from Wenner electrode spacing much Department of Computational and Applied Mathematics, Rice
University, August 2002
greater than L. Also, larger electrode distances make possible
to overcome any local variations of resistivity of the ground.

Copyright 2014 IEICE


792

Das könnte Ihnen auch gefallen