To improve the prediction accuracy of real flow in wireless network, a new prediction algorithm, state prediction algorithm was proposed basing on ARMA model and Wavelet Transform(SPWA). In this algorithm, first defined distribution character and theoretical foundation of ARMA model, then yielded higher real flow prediction accuracy basing combined predictions of ARMA and wavelet transform. at least, simulations were conducted to study the key factor of algorithm through OPNET and MATLAB. The results show that, comparing to FARIMA model and MF-FIR model, SPWA has better suitability.