Abstract: A local projective noise reduction for nonlinear time series is here introduced. First, one would reconstruct the time series using delay coordinates. In the high dimensional phase space, the local projective algorithm is used to separate it into different components in the time series by dividing the phase space into orthogonal sub-spaces. Though numerical simulations of Lorenz model, it is proved that the algorithm has an excellent effect on removing random noise in nonlinear time series. Furthermore, two algorithms of extracting weak feature signals from a time series are discussed.

Key words: Local projective, Noise reduction, Nonlinear time series, Weak feature signals