Gas Prospect Derisking in the Drava Basin: an Unbiased Approach Using Neural Networks

Abstract

Summary Derisking a prospect is a key step in exploration. Usually several characterization techniques using different types of input are run independently to validate a prospect, but it is a complex task to integrate the different results and most of those techniques and analysis rely on interpreter dependant steps. For a prospect evaluation in the Drana basin in Croatia targeting amplitude anomalies, where drilling history has proven that some anomalies are commercial and some anomalies turned out to be low saturated gas or low velocity shales, an effective workflow was needed to analyze the risk associated to the prospect. To perform this task an unsupervised classification technique has been used. This approach enables the use of several types of input altogether, removing the need for integration at the end, so both amplitude versus offset (AVO) and spectral decomposition (SD) attributes have been used as input to the workflow to properly differentiate the different types of anomalies and highlight a class corresponding to the gas sands only. The classification result focuses on anomalies without any explicit relationship, removing interpreter dependant steps, so a refined prospect area with higher probability of success has been identified.