Use of neural networks for studying nontraditional hydrocarbon reservoirs (an example of Visean black shales of the Dnipro-Donets Basin)

• Date: 2025
• Main Authors: Kurtyi, V.O., Verpakhovska, O.O.
• Format: Article
• Language: Ukrainian
• Published: S. Subbotin Institute of Geophysics of the NAS of Ukraine 2025
• Subjects: unconventional hydrocarbon reservoirs; Dnieper–Donets Basin; petrophysical properties; organic-matter content (TOC); neural networks
• Online Access: https://journals.uran.ua/geofizicheskiy/article/view/341701
• Journal Title: Geofizicheskiy Zhurnal

The paper presents a study of Visean-stage clay shales of the Dnieper-Donets Basin using a neural network algorithm. The Dnieper-Donets Basin is among Ukraine’s prospective regions for shale gas exploration. Given the need to boost hydrocarbon production from depleted fields through non-traditional approaches, detailed geological and geophysical characterization of shale-gas-bearing strata is both timely and promising. The oil and gas potential of combustible shales is largely governed by the content of organic matter, specifically the total organic carbon. For estimating total organic carbon in organic-rich rocks from wireline logs, the Passey method is widely used. We propose a new approach to forecasting organic-matter content in the target intervals when only a limited suite of logs and a restricted core dataset are available. The approach leverages state-of-the-art techniques, namely, a neural-network algorithm. Rapid advances in neural networks have encouraged their uptake in geophysical workflows, especially where input data are sparse. For this work, we employed a three-layer neural network of the multilayer Perceptron type, which directly maps inputs to outputs through successive neuron layers. We demonstrate that combining the common Passey technique with a neural-network algorithm not only yields sufficiently accurate predictions of organic-matter content within shale intervals but also refine previously obtained results.