Ground Resistance Estimation Using Inductive Machine Learning

Summary

This work demonstrates the application of entropy information based inductive learning techniques for the estimation of the ground resistance of grounding systems, used for the safe operation of electrical installations, substations and power transmission lines. Ground resistance value must be kept in low levels, so that grounding systems are able to provide the lowest impede path for fault currents to be dispersed into the earth, in the shortest possible time. The key concern of the present work is the evaluation of grounding systems performance and the possible estimation of future formed values of ground resistance, since soil resistivity and rainfall data are available. For this purpose, measurements of soil resistivity in various depths of the ground and of rainfall height have been carried out over a period of four years or so, in a particular field inside the university campus. At the same time, the ground resistance values of few grounding rods, encased in ground enhancing compounds, are recorded as a function of time. The computational method generalizes over numerical data corresponding to these ground resistance measurements. For the modeling of the data, classes are represented by discrete intervals of measurements. Decision trees are constructed for approximating the discrete-valued target function of ground resistance and then they are represented by production rules in order to improve the model comprehensibility. The error rates and the performance of the model on unseen cases are determined by a v-fold cross validation approach. Results seem promising for further development of the method. Inductive machine learning is used not primarily as a classifier aiming at obtaining high accuracy, but more as a knowledge discovery tool, finding interesting rules and decision patterns of high quality, to be checked further with statistical techniques.

Additional informations

Authors

Reuter, Templeton, McLeod, Bruintjies