GROUND’2010 & 4 th LPE International Conference on Grounding and Earthing & 4 th International Conference on Lightning Physics and Effects Salvador - Brazil November, 2010 AN ARTIFICIAL NEURAL NETWORK FOR ESTIMATING THE GROUND RESISTANCE Fani E. Asimakopoulou 1 George J. Tsekouras 2 Ioannis F. Gonos 1 A.X. Moronis 3 Ioannis A. Stathopulos 1 1 High Voltage Laboratory, National Technical University of Athens, Greece 2 Departement of Electrical Engineering and Computer Science, Hellenic Naval Academy, Piraeus- Athens, Greece 3 Department of Energy Technology, Technological Educational Institute of Athens, Egaleo- Athens, Greece Abstract - The objective of this paper is the development of an artificial neural network model based on resistivity measurements and weather conditions for the prediction of the ground resistance. On that purpose extensive experiments were carried out and the resistivity of the soil was measured. Then, an artificial neural network has been applied on the experimental data for the estimation of the ground resistance. The correlation being achieved between the measured and the estimated values of the ground resistance is more than satisfactory. 1 - INTRODUCTION The purpose of any grounding system is to provide a path of low resistance to faults and lightning currents and to assure the protection of any person in the vicinity of the grounded facilities [1]. The value of the ground resistance greatly depends on the grounding system and the properties of the soil, where the system is embedded. The earth resistivity, in turn, varies with temperature, moisture, salt content and compactness of the soil [2], [3], [4], [5], [6]. Since these parameters vary during the year, the grounding system cannot be characterized by a single value of resistance. So far several approaches, which are based on empirical and approximating equations, have been developed for the estimation of the ground resistance. The approach proposed in this paper, takes advantage of the capability of the artificial neural networks (ANNs) to recognize relationships among quantities, that otherwise would be difficult to be modeled. So far, ANNs have been successfully used by Salam et al. [7] for modeling and predicting the relationship between the length of the buried electrode and the grounding resistance. Amaral et al. [8] have implemented an ANN approach in order to map the relationship among the soil resistivity, grounding resistance, frequency, and current peak. In this paper an ANN has been trained and validated by using experimental data of ground resistivity and rainfall in order to predict the ground resistance. 2 – BASIC CONSIDERATION 2.1 – SOIL RESISTIVITY MEASUREMENTS Many techniques have been developed for the measurement of soil resistivity. The most important of them are: a) the Wenner method, b) the Schlumberger method, c) the dipole method and d) the alternate configuration method. The Four-Point (Wenner) method [2] is the most accurate for the determination of the average soil resistivity. The measurements of the soil resistivity [3] were conducted in the area of Athens from October up to July, whereas the meteorological data were provided by the National Meteorological Authority of Hellas. In Fig. 1 the rainfall data are presented. The measurements of soil resistivity were conducted according to the Wenner method by using the NORMA 1805 GB 2D/E ground tester. Figure 1- Rainfall (mm) As shown in Fig. 2 four electrodes 45cm in length are driven in line, in a depth b at equal distances α from each other. A test current (Ι) is injected at the two terminal electrodes and the potential (V) between the two middle electrodes is measured. The ratio V/I gives the apparent resistance R (in Ohms). The apparent soil resistivity ρ is given by the following formula [2]: 4 2 1 2 2 2 2 4 aR a a a b a b π ρ ⋅ ⋅ ⋅ = ⋅ + − + ⋅ + (1) If a>>b the apparent resistivity is calculated by: ρ π = ⋅ ⋅ ⋅ 2 Ra (2) Measurements have been carried out on a 40m line at spacings between the electrodes of 1, 2, 3, 4, 5, 6, 8, and 10m [3]. The ground resistance is measured according to the fall of potential method and the 62% rule [2]. The distance between the current electrode and the electrode being tested (1.5m long) is 40m, while the potential electrode is placed 24m away from the electrode under test (Fig. 3).