#TITLE_ALTERNATIVE#

• Main Author: Robby Rizaldi (NIM 124 03 003), Agi
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/7057
• Institution: Institut Teknologi Bandung

Abstract: <br /> <br /> <br /> <br /> <br /> <br /> <br /> Deconvolution is a seismic method which has purpose in eliminating the effect of seismic wavelet, in order to get the estimation of sub surface reflectivity. The elimination of wavelet effect will increase seismic temporal resolution, and this will give an easier way for making an interpretation of seismic section. Besides that, we could use the results of reflectivity estimation as a better input for seismic inversion process. In this research, we will use neural network as a method in the deconvolution process. <br /> <br /> <br /> <br /> <br /> <br /> <br /> For the deconvolution process itself, we will use backpropagation as an application of neural network. Backpropagation is one of a computation program that applied neural network, in order to solve a non linier problem. In this neural network, we used an input layer, three hidden layer, and an output layer. As a transfer function, we used logsig mode for hidden layer and linier mode for the output layer. <br /> <br /> <br /> <br /> <br /> <br /> <br /> A synthetic model was defined as a wedge model with four layers where every layer has a different velocity value. On the other hand, we will also use a clean synthetic data (without noise) and a synthetic data with noise that gradually increase from 5 until 25 percent noise. In this research we used a random noise, and furthermore we could call it as a Color Noise, because it has been applied through a butterworth filter. <br /> <br /> <br /> <br /> <br /> <br /> <br /> Deconvolution process using neural network method shows an optimum results in describing layer boundaries, especially for minimum noisy data. We could have a good deconvolution operator filter from a little less noisy data, so it can be applied for a low percentage noisy data. Otherwise, a high noisy data will make the deconvolution operator filter work worse. <br />