Predicting the pile static load test using backpropagation neural network and generalized regression neural network – a comparative study

Abstract

In this paper, two neural network models; the Back Propagation Network, and the Generalized Regression Neural Network were built to predict the downward displacement of the entire static load test for Continuous Flight Auger piles. The models aim to increase our ability to predict the CFA pile’s performance which depends on parameters such as the ground properties and the pile’s characteristics. These parameters along with others were selected as input for a sample of more than six thousand field data points to train the two models. The models were then tested using three independent field tests and the results were matched against the actual measurements. The results showed that both models can predict the two loading-unloading cycles. To compare between the two models five statistical measures were used to show that the Back Propagation approach was more accurate than the Generalized Regression Neural Network while the latter was more precise.