Before an oil producer begins a decommissioning programme, it must identify potential pay zones within its existing infrastructure where hydrocarbons could still be extracted economically. It also needs to prioritise new areas for exploration research, something which can take months to achieve.
With a remit to help the industry identify new technologies to improve and optimise oil exploration and production, the Oil and Gas Technology Centre (OGTC) wanted to use a data-driven approach to find a better way of identifying overlooked pay zones. It approached Merkle to help utilise the vast amount of data in the Oil and Gas Authorities national data repository (NDR) to provide a solution.
Merkle’s approach was to focus on oil well log data in the NDR for each site of the nine industry sponsors. However, with no formal definition of what ’’pay’’ meant, and past economic conditions impacting on wells not having fulfilled their financial potential, Merkle needed to develop a standard definition.
Using a cited research paper, Merkle created a standard definition on pay the operators could work back from and could easily interpret to meet their investment criteria to determine if drilling would be financially viable. This ensured the credibility of the outputs once they were deployed.
One goal for the project was to use the data to find new wells for investigation. However, the data available was biased as most successfully-tapped areas were in regional clusters, meaning the data was heavily skewed to locations where there were already hydrocarbons.
To unbias the data, Merkel built a semi-supervised model with the understanding that there would be overfitting on to it. It also developed a complementary model that was unsupervised and unbiased to allow a better prediction of new regions to explore. The combination of these models helped to uncover new potential areas of interest.
With 150 million data points, there was a challenge around addressing the sheer volume of predictions. To overcome this, Merkle built a cluster analysis using the unsupervised model and third-party seismic data to reduce the outputs by grouping them into areas of interest.
The analysis identified which cluster was associated with the prediction by overlaying the semi-supervised prediction of a pay zone. Using the analysis with the two models, areas could be identified and then more precise hydrocarbon areas defined for investigation.
The data was bought to life using a data visualisation tool, creating an interactive scatterplot of all the wells in the data set. Analysis on this scale had never been seen in the oil and gas industry before and has taken the guesswork out of the process. Geo-physicists and petro-physicists can now identify and explore different areas in more detail in a fraction of the time it would have taken previously.