Design of ground improvement measures against consolidation settlement via visualizing latent space

Abstract

For road embankment on soft clay ground, ground improvement will be implemented to prevent consolidation settlement. Machine learning is performed using the ground improvement pattern and the embankment deformation pattern obtained by the genetic algorithm method. By using two kinds of compounders composed of the obtained convolutional neural network, it is possible to predict the improved pattern and the embankment deformation pattern from the two-dimensional latent space. Next, a two-dimensional latent space is divided into blocks, and a prediction score whose value increases when the improvement effect is high is defined. When the contour of each block of the average improvement pattern and the contour of the prediction score are displayed and compared in a two-dimensional latent space, a ground improvement pattern having a high improvement effect can be easily designed.