Аннотация:There are two methods to earthquake early warning system: the method based on a network of
seismic stations and the single-sensor method. Both have advantages and drawbacks. The current
systems rely on high density seismic networks. Attempts at implementing techniques based on
the single-station principle encounter difficulties in the identification of earthquake in noise. The
noise may be very diverse, from stationary to impulsive. It seems a promising line of research to
develop hybrid warning systems with single-sensors being incorporated in the overall early
warning network. This will permit using all advantages and will help reduce the radius of the
hazardous zone where no earthquake warning can be produced. The main problems are
highlighted and the solutions of these are discussed. The system is implemented to include three
detection processes in parallel. The first is based on the study of the co-occurrence matrix of the
signal wavelet transform. The second consists in using the method of a change point in a random
process and signal detection in a moving time window. The third uses artificial neural networks.
Further, applying a decision rule out the final earthquake detection is carried out and estimate its
reliability.