Application of hybrid deep transfer learning approach in nuclear oil well logging technique

Automated interpretation of nuclear well-logging data is critical for accurate hydrocarbon reservoir characterization but remains challenged by data scarcity and interpretive complexity. In this study, we propose a hybrid deep transfer learning architecture that integrates one-dimensional Convolutional Neural Networks (۱D-CNNs) and Bidirectional Long Short-Term Memory (Bi-LSTM) layers to process spectral signatures from dual-detector configurations, specifically High-Purity Germanium (HPGe) and Thallium-doped Cesium Iodide (CSI(T۱)). By leveraging MCNPX-based synthetic simulations for pre-training and applying transfer learning to field spectra, the model effectively mitigates the domain shift between synthetic and real-world data. The proposed framework demonstrates superior performance in discriminating between hydrocarbon-bearing and non-bearing zones, achieving a classification accuracy of ۹۶.۸%. Notably, this hybrid dual-detector approach provides an improvement of ۲.۶% over HPGe-only and ۷.۳% over CSI(TI)-only configurations, confirming the efficacy of the proposed multi-detector fusion and transfer learning strategy for robust well-logging interpretation.

کلیدواژه ها:

spectral analysis ، deep transfer learning ، HPGe detector ، CSI(T۱) detector ، nuclear well logging