Research on RBF neural network method of singularity detection in chaotic time series

This paper researches on using RBF neural network for singularity detection in chaotic time series. History data are used for off-line training to RBF network, and then comparing the result of output between actual and desire, if the difference of their values is above a certain threshold, the data will be judged singular. Parameters of RBF network will be renewed according to the real-time data. The raw data produced by Lorenz system, the data with disturbance and the measured-data from oil pipeline pressure are used to test capacity of RBF network for singular signal anti-interference, weak signal examining and multi-step forecasting respectively. The research conclusion shows that RBF network not only has a strong ability for detecting faint signal in chaotic time series, but also has a good effect on singularity detection in measured-data.