Neural Net Expansion Model for Fissured Strong Expansive Soil

Abstract

Fissured strong expansive soil swelling behavior is complicated. In this paper, considering the typical filling fissures of strong expansive soils, fissure rate K r was given as a fissure content quantitative indicator. A prediction model was developed for the prediction of swelling effect on a fissured strong expansive soil using BP neural network approach, the gradient descent and the conjugate gradient algorithm methods were adopted. The actual test and predicted results of the two algorithms network showed high degree of similarity. The BP neural network model described by fissure rate, dry density, initial moisture content and overlying load can meet the precision requirements. The conjugate gradient method when compared with the gradient descent method, has a significantly improved calculation efficiency, the convergence rate is about 30 times lesser than the latter, therefore, conjugate gradient algorithm BP network prediction model for swelling in the actual engineering calculation has obvious advantages.