Perm Algorithms Advance Mineral Exploration
Before drilling a borehole, geologists gather information about the ground: what types of rock are present, how they are arranged, and what has happened to them over hundreds of years. The resulting data are vast and highly varied, ranging from text and descriptive records to numerical measurements. Neural networks being trained to find ore deposits struggle to process this volume of information. Doing so requires substantial computing time and memory.
Researchers at Perm National Research Polytechnic University (PNRPU) have developed a “translator” that converts geological language into numbers. They took complex descriptions, including rock names, changes in rock formations and various geological indicators, and turned them into a compact set of numerical values that a neural network can process. The amount of data has shrunk, while the most important information remains intact.
Less Memory, More Ore
Instead of 13 features, the AI model uses just six, cutting memory requirements in half and training time by one-third. The system no longer learns from a large collection of heterogeneous data; it works with a condensed “concentrate” of the information. Even so, its accuracy on test data reaches 88%.
On a geological map, the algorithm classifies areas by their exploration potential: high, medium, low or zero. Geologists can see where deploying a drilling rig makes sense and where it does not. This approach could save mining companies billions of rubles by reducing the number of costly dry holes. The method has already been field-tested using real data from the Agbau gold deposit complex in Côte d’Ivoire. The software is written in Python, making it flexible and straightforward to integrate into other systems.
Technology Without Borders
Russia continues to deploy AI tools in geology, from diamond and oil exploration to mineral identification and the optimization of exploration strategies. ALROSA, a diamond mining company, for example, has developed an AI system that analyzes more than 50 years of accumulated geophysical data to identify promising areas for exploration, including sites with complex geological structures.
Gazprom Neft used AI to identify additional hydrocarbon deposits at the Vakhskoye field in the Khanty-Mansi Autonomous Area and Tomsk Region. Preliminary estimates put recoverable reserves at no less than 100,000 metric tons of oil. The system analyzed data from operating wells and identified promising formations that had previously gone undetected.
Geological models of deposits generally account for the concentration of valuable components and the volume of reserves, but they rarely consider how amenable the ore is to processing – that is, how easy or difficult it will be to extract metals and minerals at a processing plant. Work by researchers at the Kola Science Centre of the Russian Academy of Sciences and Murmansk Arctic University addresses this gap. Their algorithm produces a precise mathematical forecast of which exploration strategy and processing or beneficiation method will be most effective for a specific rock sample.
Researchers at Moscow State University, meanwhile, have developed software that automatically identifies minerals from photographs of polished ore samples. The tool is designed to accelerate mineralogical studies. It achieves 91.7% recognition accuracy on the LumenStone test dataset.
Promising Accuracy
PNRPU previously presented a software method that combines AI, geoinformation technologies and spatial analysis. The system identifies areas with high gold potential with 90% accuracy and areas with medium potential with 89% accuracy. Those tests also used data from a deposit in Côte d’Ivoire. The latest development can therefore be seen as a continuation of that technological work.
Russia’s mining industry operates in some of the world’s most challenging conditions. By reducing computational requirements, the PNRPU algorithm could become an important tool for exploration in remote regions such as Siberia, the Arctic and the Russian Far East. It could also become part of digital platforms like the one being developed in Zabaykalsky Krai, where developers have already announced plans to cut subsoil users’ costs by 25%.
PNRPU’s new technology does not merely halve the entire mineral exploration cycle. Instead, it makes the computational core of exploration faster and less expensive, providing a foundation for more efficient forecasting systems. Russia still has vast, unexplored areas with potential for new discoveries. Last year, 317 new deposits of solid minerals and hydrocarbons were added to the state register of mineral reserves. Smart technologies could help increase that number further.
