Density Cube Prediction in Carbonate Layer to Detect Dolomitization Based on Deep Neural Networks
P.F. Popova, I.I. Priezzhev and S.V. Egorov
Event name: Geomodel 2017
Session: Кинематическая и динамическая интерпретация данных сейсморазведки (атрибутный анализ, все виды инверсии и т.п.)
Publication date: 11 September 2017
Info: Abstract, PDF ( 1.09Mb )
In presentation shows example density cube prediction in carbonates in Caspian sea area to detect dolomitization zones using deep neural network. The prediction is based on full stack seismic cube and several well logs. To train deep neural networks we used hybrid training technique based on a combination: deep neural network input generalization for first layers, genetic algorithms together with a gradient methods and Tikhonov regularization. Using deep neural network allow to build prediction operator with high level of the power of freedom and as result good quality of the prediction.