Acceleration
Response Spectra (ARS) for mining tremors in the Upper Silesian Coalfield are
generated using neural
networks trained by means of Kalman filtering, applying
DEKF (Decoupled Extended Kalman Filter) algorithm. A comparison of ARS
predicted by DEKF and by standard networks, trained by the Rprop (Resilient-Peropagation)
learning method is presented. Due to using of time delay input the
results of
DEKF and Rprop applications give predictions of ARS with a great accuracy.
Moreover, it can be stated that the application of Rprop gives
approximation from above of measured of ARS and in many parts of ARS the approximation by
DEKF is from below. These results are also compared with the approximation
obtained in <2>, where results were worse because of using other inputs.