Prediction of vertical displacements in civil structures using artificial neural networks

Abstract

This article attempts to analyse and predict vertical displacements of measurement-and-control network points located on civil structures founded on expansive soils, using artificial neural networks. Geodetic monitoring of civil structures consists in regular measurements of control point networks and interpretation of results. The obtained values of displacement provide sets of significant data which enable determination of the influence of changes in ground water conditions of the subsoil on the deformation processes occurring in structures founded on it. Using such data sets, it is possible to draw conclusions regarding the dynamics of the occurrence of deformation and to develop a geometric model of displacements. In recent years, methods of prediction based on artificial intelligence have been increasingly prominent. Neural networks and evolutionary algorithms, which can supplement each other, make advanced tools applied in the process of prediction of deformations. In order to forecast displacements of control points, demonstrating changes in a civil structure, multi layer artificial neural networks are employed in this article, taught using the method of error backpropagation and gradient optimization methods. The analysed results in the form of height differences were obtained through a series of measurements on a civil structure, taken by means of precise levelling at monthly intervals.