||The paper is devoted on the problem of real-time earthquake prediction. An approach for realtime prognoses, based on classification algorithm of strong motion waves with neural network and fuzzy logic models is suggesting. As input information for the neural network, build with Kochonen learning rules, are given the parameters of recorded part of accelerogram, principle axis transformation and spectral characteristics of the wave. With the help of stochastic long-range dependence time series analysis is determined the beginning of destructive phase of strong motion acceleration. Developed seismic waves classification gives possibility to determine different kind prognoses models for different king of classified waves. The prognoses of destructive seismic waves are realized with learning vector quantization and self-organizing map.