Abstract: Because of the influence of the noise contained in vibration signal，the nonlinear characteristic parameters of flue gas turbine are usually extracted incorrectly.In order to eliminate the noise，the wavelet threshold de-noising is applied in this paper.Firstly，the algorithm of this method is introduced.It is based on the theory of wavelet transform.The threshold and threshold function are selected and improved.By wavelet decomposition，the wavelet coefficients of the original signal and the noise on different scales are obviously different.According to this，the scale of decomposition can be chosen properly.Secondly，this method is tested by the signal of Lorenz attractor and the flue gas turbine signal.The result shows that this method can reduce the noise effectively and the local feature of the original signal is well reserved.Compared with the pseudo-phase portrait reconstructed from signal contained noise，the pseudo-phase portrait reconstructed after the wavelet threshold de-noising is more regular.By this method，the correlation dimension，which can reflect different fault condition for nonlinear system，is estimated accurately.It is proved that this method can improve the accuracy of signal analysis.

Key words: flue gas turbine, signal analysis, wavelet threshold de-noising, correlation dimension, pseudo-phase portrait