Robust Time Series Prediction of Neural Network
He Pi · Kongzhi yu juece · 2001
To solve the problem of the robust prediction of neural networks, the paper proposed a universal method of nonlinear model identification. The method is based on the nonlinear partial autocorrelation. It could determine the order of nonlinear auto regression by investigating the irreducible dependence between current quantities of time series and high order historical quantities. Computer simulations perfectly supported our idea.