The impulse response of BP neural networks and its application to seismic wavelet extraction

Artificial neural networks (ANNs) are increasingly being applied in geophysical data interpretation largely due to the fact that they have been shown to be universal function approximators. However, as ANNs act like “black boxes”, there is concern about their reliability. An understanding of the learning of BP neural networks for certain kinds of function approximation can be archived by utilizing the concepts of impulse response from the signal theory. This naturally leads to an algorithm for seismic wavelet extraction constrained by well information. This algorithm is verified with synthetic and real data