Log Prope Lorties Prediction from Seismic Data using the Multi Attribute Transforms. A case study of Algerian Field : JournéesThématiquessur la Prospection Sismique 24 & 25 Septembre 2019

Abstract

In this paper, we describe a new method for predicting well logging properties from seismic data. A 3D seismic volume, obtained from Software (Petrel), after analysis can provide a set of logs of a well characterizing this seismic volume. The logs recorded in this well are diverse and numerous. The prediction of porosity from log data remains one of the best information to obtain. However, the 3D seismic volume allows us to calculate some attributes based on samples. The multi-attribute transformation is a linear or nonlinear transformation between attributes and values selected as Inputs values. A step-by-step regression process, which derives subsets of larger and larger attributes, is then applied. Just as a convolution operator is used to solve the frequency differences between the desired values and the seismic data. A feed forward multi-layer network (MLFN) and a probabilistic neural network (PNN) are then applied to this system to estimate the parameters required by the system in which the probabilistic neural networks are the best responders. The application is made for real data. The prediction of the parameters allows us to observe a continuous improvement of the predictive power which seems significant both on the learning data and on the validation da