Using artificial neural networks for the construction of contour maps of thermal conductivity

Abstract

In this paper a neural network is used for the construction of a contour map. The particular case of the thermal conductivity map of the ground of the island of Cyprus is considered, with archived data at a number of boreholes throughout Cyprus being used for training a suitable artificial neural network. The data were randomly divided into a training and a validation dataset for a multiple hidden layer feed-forward architecture. The correlation coefficient obtained between the predicted and the training dataset is 0.966, indicating an accurate mapping of the data, while the validation (unknown) dataset exhibits an also satisfactory correlation coefficient of 0.955. The dataset was broadened by embedding the patterns used for the validation into the training dataset with the correlation coefficient equalling a higher 0.972. The available input parameters were then recorded for each grid point on a detailed topographic map of Cyprus, whereby the neural network was used to predict the thermal conductivity at each point. The coordinates and the estimated conductivity were then used as input to a specialized contour drawing software in order to draw the geothermal contour map.