Data processing of reflection seismic data by use of neural network

Abstract

The present paper proposes an algorithm for data processing of reflection seismic data using of neural networks. A neural network algorithm was applied to the reading of arrival time of first break signal, the recognition of waveform in seismic trace and the automatic picking of result of constant velocity scan among the various data processing techniques. A layered network with the correct answer, so called, teacher's signal, in training period by the error back propagation algorithm was used. The general procedure of processing of reflection seismic data by use of neural network is as follows. 1. 1. Constitution of the most suitable network for the target processing. 2. 2. Setting of the weight values to all units in the layers and the teacher's signal. 3. 3. Calculation of the output signals from the output layer by activating the network. 4. 4. Estimation of learning signal from the energy of errors between the actual output signal and the teacher's signal. 5. 5. Calculation of the change of weight values by using learning signal so as to minimize the energy of errors between the actual signal and the teacher's signal. 6. 6. Steps (3) to (5) are repeated till the errors fall into the designated limitation or the designated learning count is reached. As a result of model studies, it was determined that the proposed algorithm performed the readings of arrival time of first break signal, the waveform recognition and the automatic picking of velocity analysis result with good accuracy.