engleski

Successfulness of different neural network algorithms for missing well log data prediction – Example from the Sava Depression

Intervals with missing well log data can be successfully amended with neural network predicted data (Sagaff and Nebrija, 2003). Process for completing well log data consists of training the neural network on the well with the complete set of curves and applying the neural network to predict the missing data. For this procedure well log curves, least dependent on mud properties, were chosen for better well to well to well prediction. Three wells with gamma ray, neutron porosity and acoustic well log curves were selected form Kloštar oil field. Three different neural network types were used: multi layer perceptron, radial basis function and generalized regression neural network. Program used for the neural network analysis was StatSoft STATISTICA 7. Well to well prediction was successfully achieved. Best results came from multi layer perceptron and generalized regression neural networks.

neural networks; well logs; Pannonian Basin

nije evidentirano