Pile Samples Classification Method Based on the Self-Organizing Map Neural Network

To reduce the noise in learning samples while using BP neural network to predict the bearing capacity of pile foundation, a self-organizing map neural network was adopted to classify the collected pile samples in the paper. Firstly, to maintain the SOM network at a stable situation, pile samples were discriminated into symbol codes and character codes, and a new coding model of pile character was established, by which a SOM neural network's weight formula of reduction was derived. Then, clustering of pile samples were shown by calibrating the maximum response cell of the self-organizing map neural network. Finally, case studies using the clustered samples as input vector to a BP network were presented, and the results showed that it was a good alternative approach for estimating the bearing capacity of pile foundation by using the improved solution with the characters of simplicity.