Abstract
Manual seismic interpretation has long been the backbone of E&P workflows however it is a process that demands significant time and a highly skilled interpreter. Artificial intelligence (AI) technology, specifically those that utilise deep learning convolutional neural networks have emerged as a credible alternative, producing results of comparable accuracy and detail.
A key area where AI solutions have already accelerated seismic interpretation workflows are in the delineation and mapping of faults and horizons, as well as identifying and defining features, such as karst platforms or injectites. These workflows have been tested on datasets of varying qualities and ages, as well as on both onshore and offshore datasets.
Once the faults can be detected by the AI networks, the next stage becomes horizon detection and this is done by identifying every single peak and trough in a seismic cube however, accurately correlating individual horizon patches proved to be a much bigger challenge than detecting the discontinuities. A new solution was developed that is entirely aware of faults and understands the interplay between the faults and the horizons it is analysing. The fact that the horizons are aware of the faults means these objects are the perfect input to generate a watertight model. The latest stage of the AI seismic interpretation workflow involved creating a solution to identify and extract geological features such as channels and injectites. AI networks enable the fast, accurate and unbiased analysis of dozens of volumes, applying the same criteria and rigour to every voxel in every volume. For example, every partial stack – angle stacks, azimuth stacks – can now be processed with the same AI network, allowing the human interpreter to focus on comparing and interrogating the resulting AI fault volumes to understand their significance. Furthermore, due to the AI Network's ability to assign more nuanced fault confidence values from 0 to 100%, we can investigate the uncertainty around the structural model based on fault confidence results.
Finally, all of these AI Networks can be fine-tuned by the addition of manually interpreted data labels to add extra information to the neural networks. This can result in increased fault recognition, reduced false positives and an increase in fault confidence amongst many other desirable results. These AI-based techniques significantly enhance the speed and quality of the interpretation of the subsurface. With the help of AI we can rapidly identify and extract faults, horizons and geobodies with unprecedented detail and accuracy. The time saved on manual interpretation can be used to better understand the geology and analyse feasible scenarios, enabling regional and field assessment to benefit from the most up to date information and ensuring high-quality exploration and development decisions.
