SUBSTITUTIONAL MODEL OF THE SOIL BASED ON ARTIFICAL NEURAL NETWORKS

Abstract

This paper presents an interpretation of the results of measurement of specific soil resistivity by means of artificial neural networks. The model based on artificial neural networks replaces the soil which can be physically considered a two-layer medium with a vertical change of the specific electric resistivity and a horizontal boundary line between the layers. Learning of the neural network was performed under supervision using the input dataset obtained by means of a very accurate theoretical model of the double-layer soil. The proposed algorithm that approximates non-linear soil properties using the artificial neural network is reliable in assessment of the soil parameters and specific electric soil resistivity. Application of the substitutional model of the soil based on neural networks is demonstrated by a realistic example, determination of parameters of the double-layer soil from the measured data obtained by the Wenner technique for measuring the specific soil resistivity. For simplicity of presentation and model comparability, the current probes (poles) are replaced by the ball electrodes, i.e. spot fi eld sources. The results obtained are analytically and graphically presented and discussed.