Abstract: Making full use of the multi-resolution characteristics in the wavelet transform,we can divide a signal in the time domain into different bands of frequency by using wavelet decomposition and reconstruction.The best scale of decomposition is decided by the minimum scale in which rms error of approximate signal is little and from this scale to another scale rms error of approximate signal is short odds. The decomposed signal is not only more simple in frequency component but also more stationary in the time domain than original signal. Auto Regressive Moving Average Model is employed to predict the decomposed and reconstructed sub-signals with different bands of frequency. Then the expected signal prediction in time domain is obtained by synthesizing these sub-signals prediction.Experimental prediction result with maximum earthquake magnitude and sum of earthquake release energy indicates that the model improves the prediction accuracy. There is a high correlativity between experimental prediction result and observations. And there is a low discrete distribution in difference between experimental prediction result and observations. This methods provides a relatively accurate forecast for earthquake situation.

Key words: Wavelet transform, ARMA model, Earthquake activity, Maximum earthquake magnitude, Forecast