Abstract: Due to the nonlinearity and complexity of deformation evolution of geotechnical engineering , it is difficult to describe it with simple mechanical and mathematical model. A method for forecasting the stress and displacement nonlinear time series is proposed based on constructing radial basis function neural network using particle swarm optimization algorithm .After determination of units’ number in RBF layer using k-means, all parameters such as central position, shape parameter and weights of RBFNN are estimated dynamically in global with particle swarm optimization. The engineering case studies reveal that this model has high accuracy and a good prospect for nonlinear time series forecasting of geotechnical engineering.

Key words: particle swarm optimization, radial basis function neural networks, geotechnical engineering, nonlinear time series, forecast