--> Abstract: Fracture Identification from Well Logs Using Neural Networks and Its Application to Fracture Production of a Chalk Reservoir, by Guohai Jin; #90914(2000)

Datapages, Inc.Print this page

Guohai Jin1
(1) The University of Alabama, Tuscaloosa, AL

Abstract: Fracture Identification from Well Logs Using Neural Networks and Its Application to Fracture Production of a Chalk Reservoir

Fracture identification from conventional well logs is difficult, since no single log response is completely diagnostic of fracturing. The drawbacks of the traditional approach are less efficiency, interpreter bias, and lack of quantitative accuracy. To overcome these difficulties, I have employed a back propagation neural network (BPNN) which utilizes multiple logs for identification. The BPNN has an input layer containing several input nodes (neurons), and output layer containing one node, and one hidden layer with four nodes. The nodes in different layers are interconnected by weights, which are updated by supervised training. The training process was accomplished by a number of training samples form pilot wells which typical log responses indicative of the presence or absence of fracture at some depth. The criteria for detecting fractures in chalk are (1) noticeable separation between resistivity logs; (2) caliper expansion; (3) lower sonic amplitude. The output of the network is a fracture index (FI) log ranging from 0 to 1 to describe the probability of fracture existence. The algorithm was applied to a fractured chalk reservoir developed in the Gilbertown oil field, located in a rider block of the Gilbertown graben system which is deformed into a drag fold, southwest Alabama. Continuous FI logs are calculated for all the wells in the field area using the trained network. It can be seen that zones of high FI are confined within the drag zone. The production intervals correlate well with the high fracture index zones. A column of fracture production from the chalk is further defined based on the calculated fractures and drag geometry. It is concluded that the effectiveness of fracture index by artificial neural networks provides better characterization of fractured reservoir properties over the conventional method. This technique can also be applied to other areas with similar lithology.

AAPG Search and Discovery Article #90914©2000 AAPG Annual Convention, New Orleans, Louisiana