Artificial neural networksbased-model for forecasting critical height of GRW

This paper presents an artificial neural networksbased approach for predicting the critical height of GRW. Seven major affecting factors have been used for analyzing the general failure cause. A radial basis function neural network (RBFN), as well as a back propagation neural network (BPN) for comparison, is trained and tested using 23 series of centrifuge model test data, 2 fullscale test data, and prototype date of a practical project. The modeling results indicated that the RBFN is much better than the BPN on learning speed, prediction accuracy and generalization ability. The paper provides a reference for GRW design.