EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Deep Learning Seismic Impedance Inversion with Adaptive Fusion of High-Frequency Well Priors

Mingxuan Li, Cong Niu, Suzhen Shi, Jianxiang Feng

Applied Sciences · 13.11.2026 · 16 (22) · 11284

DOI: 10.3390/app162211284 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 13.11.2026. Verfasst von Mingxuan Li und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, CNOOC Research Institute Ltd., Key Program of the National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

High-resolution seismic impedance inversion is important for characterizing reservoir heterogeneity and fine stratigraphic structure, but limited seismic bandwidth and sparse well control hinder high-frequency recovery. We propose a deep learning method that adaptively fuses high-frequency well priors into impedance reconstruction. Horizon-derived relative geologic time guides prior propagation from training wells along strata, and well distances and seismic structural similarity determine prior contributions and well support weights. Seismic data, the prior, well support weights, local waveform similarity, and RGT and coordinates form the inputs to a one-dimensional convolutional network, which predicts a mid-frequency impedance increment and high-frequency control weights. The control weights regulate the prior contribution and are supervised by the ratio between measured and propagated high-frequency amplitudes at the training wells. The increment and weighted prior are added to a low-frequency background to reconstruct high-resolution impedance. Training combines sparse well supervision, seismic forward modeling, lateral stratigraphic constraints, and amplitude-ratio supervision. In synthetic tests, the method delineates thin beds more clearly than conventional deep learning methods, and the supervised weights raise high-frequency correlation from 0.760 to 0.835. On a field section, inverted impedance matches validation-well measurements better than total variation inversion.

Förderung

  • National Natural Science Foundation of China
  • CNOOC Research Institute Ltd.
  • Key Program of the National Natural Science Foundation of China

Zitieren

APA

Li, M., Niu, C., Shi, S., & Feng, J. (2026). Deep Learning Seismic Impedance Inversion with Adaptive Fusion of High-Frequency Well Priors. Applied Sciences, 16(22), 11284. https://doi.org/10.3390/app162211284

Vancouver

Li M, Niu C, Shi S, Feng J. Deep Learning Seismic Impedance Inversion with Adaptive Fusion of High-Frequency Well Priors. Applied Sciences. 2026;16(22):11284. doi:10.3390/app162211284

BibTeX
@article{Li2026,
  author = {Li, Mingxuan and Niu, Cong and Shi, Suzhen and Feng, Jianxiang},
  title = {Deep Learning Seismic Impedance Inversion with Adaptive Fusion of High-Frequency Well Priors},
  journal = {Applied Sciences},
  year = {2026},
  volume = {16},
  number = {22},
  pages = {11284},
  doi = {10.3390/app162211284}
}
Kostenlos starten →