Estimation of mining tremor occurrence by using neural networks

Changes of the primary strain-stress state (caused by interaction between natural conditions and mining activity) can result, under special circumstances, to the origin of seismic induced events. The question of induced seismic activity prediction was treated as a problem of lime series extrapolation of maximum cumulative amplitudes and numbers of seismic events recorded per day. The treatment was carried out by means of Multilayered Perceptron Neural Networks (MLP NN). The application to mining tremor prediction has been tested and methodological conditions have been obtained. It was proved that the prediction of the number of mining tremors per day is more precise than the prediction of future energy (maximum amplitudes). Further advance, based on the processing of seismo-acoustic activity series, is introduced.