The Thickness Prediction of the Presalt Carbonate Reservoir Based on Machine Learning Multi-attribute Seismic Technology

Seismic single attribute to predict reservoir’s parameters can’t satisfy demand for pre-salt carbonate reservoir interbedded with some igneous rocks, the paper attempts to predict reservoir parameters using seismic multi-attributes analysis technologies for improving accuracy and reliability. Insensitive attributes are expelled by manual experience through the crossplots between surface attributes and reservoir parameters from drilling wells, the remaining seismic attributes or groups are the optimum combination to resolve desired questions; The rest of seismic attributes maybe be too much and needed to decrease dimension, principal component analysis and K-L transformation are used to compress seismic attributes; artificial neural network algorithm is used to attribute pattern recognition, the reservoir parameters in the whole region are quantitatively predicted or estimated. In view of some questions such as igneous rock development、pre-salt carbonate reservoir prediction difficulty in Block L offshore, pre-salt carbonate reservoir thickness is quantitatively computed with seismic multi-attributes analysis technology, It is proved that the error is less than 5.5% between results and verified wells, the prediction accuracy is obviously higher than using one single attribute method, and reservoir thickness distribution matches geologic sedimentation pattern. The practical results show that this method is feasible in both concept and practical application from the case study, which can efficiently improve the accuracy of reservoir quantitative prediction, and provide suggestions for the exploration of reservoir distribution characteristics and subsequent deployment of development well network scheme.