Physical Properties of Reservoirs Using an Artificial Neural Network Approach: Example from the Jeanne d'Arc Basin, Eastern Offshore Canada
Zehui Huang and Mark A. Williamson
Quantitative statements of the evolution of petroleum systems in the Jeanne d'Arc Basin provides a solid basis for future exploration, reservoir modelling, reservoir development and resource management. As a contribution to such statements, we have integrated physical property measurements (porosity and permeability) and well log data from the major reservoir intervals throughout the basin using an efficient backpropagation artificial neural network (BP-ANN) modified with the Marquardt algorithm. The bulk of the data are from the Avalon Formation (13 wells), the Hibernia Formation (7 wells) and the Jeanne d'Arc Formation (12 wells). After data preprocessing and training/supervising example preparation, a model for the relationship of physical property and well log respons was established with the BP-ANN technique. Test of the BP-ANN model in other Mesozoic reservoir intervals in the same basin whose data were not used as the training and supervising examples shows good agreement between the measured and predicted values. The BP-ANN model established was used to construct permeability porosity profiles in 40 wells for the Avalon Formation, 36 wells for the Hibernia Formation and 37 wells for the Jeanne d'Arc Formation from well logs. These profiles offer a more complete basis for understanding physical properties of reservoir intervals of this basin. They allow a detailed review of vertical and horizontal variations of physical properties in key areas of this basin.
AAPG Search and Discover Article #91019©1996 AAPG Convention and Exhibition 19-22 May 1996, San Diego, California