Indonesia Conference Directory

Sandstones containing 60% of oil or gas because of porosity and permeability. Conventional method required sandstones predicted with complete data. This study used neural networks to determine the depth of sandstones and predict incomplete variables well site at Sunda Strait-south area. Data preprocessing used PCA-PLS to find the most important variables that affect the output thus improving the prediction results. Multicollinearity analysis is used to determine the data compression needed. Raw data multicollinearity result showed that the multicollinearity occurs indicated with VIF over 10 and tolerance under 0.1 at CALI and SP variables. PCA-PLS analysis use to reduce data from 13 variables into six important variables namely DEPT GR RHOB NPHI ILD Peff, these results do not experience multicollinearity. Variable that predicted is ILD with best ANN multilayer perceptron showed small standard deviation and standard error results 2.85 and 0.03. Best ANN model to predict the depth of radial basis sandstone is due to produce a regression test of 0.8 based on the results of the validation of the log image.