Scientists from Saint-Petersburg State University and Moscow State University named after M.V. Lomonosov created a neural network algorithm capable to drastically simplify and facilitate oil and gas prospecting at the continental shelf. This development opens new opportunities for developing Russian off-shore fields, where the reserves are believed to be the biggest in the world.
The sphere studied by the scientists from Saint-Petersburg State University and Moscow State University is critically important for the future of Russian oil-and-gas industry. About 70% of the explored hydrocarbons reserves are concentrated in the Arctic and Russian Far East, but only 5% of them developed due to extreme climate conditions. The upper layers of the sea bottom need to be explored with high level of accuracy before building the drilling platforms — the first 100-150 meters, where potentially paleochannels of ancient rivers may be found together with soil softening zones or bowlder blocks. If such elements are not timely identified, they may constitute a serious danger for sustainability of the sea infrastructure.
Traditional methods of seismic analysis are rather efficient, but they require huge time resources. Specialists had to manually process hundreds of thousands of seismic data sets spending months on this meticulous work. To meet more or less acceptable deadlines, the geologists were forced to use only 8-10% of the collected data, which inevitably decreased the level of detail and accuracy of the results. Special difficulties occurred when small geological anomalies of complicate shape were identified.
To resolve these problems, Russian scientists proposed a conceptually new approach using the capabilities of artificial intelligence. They trained the neural network to analyze seismic data using the limited data sets at first, and then applying the model to the entire collected data set. This allowed to automate the process of dispersion curves distinction and building velocity models, which was performed manually.
Eventually, the neural network algorithm not only facilitated data processing exponentially, but improved the accuracy of the results. The scientists succeeded in building detailed 3D models of distribution of shear wave velocity with the aggregate square exceeding 2,000 square kilometers. The new technology provides for correct recovery even of complicated velocity anomalies, which is confirmed by independent seismic observations.
This breakthrough has great practical importance. Firstly, it allows for safer development of off-shore fields accurately accounting for geological risks during construction and drilling. Secondly, it significantly decreases the time for data processing. Thirdly, using 100% of the collected data instead of 8-10% provides for unprecedented level of detail of the studies.
The plan for the future is to improve the method by way of training the model based on synthetic data, which will make it even a more reliable instrument for off-shore geophysics and will help to implement the entire potential of Russia in developing Arctic and Far Eastern fields.
