Summary Accurate identification of natural fractures is essential for reservoir characterization and stimulation optimization in unconventional reservoirs. While conventional well logs provide extensive coverage, their interpretation for fracture identification remains challenging. Traditional fractal-based methods, particularly the rescaled-range (R/S) analysis combined with the finite-difference (FD) method (R/S-FD), are often compromised by the inherent limitations of the three-point second-derivative calculation, which is highly susceptible to data noise and sampling intervals, leading to unreliable fracture identification, especially in long horizontal well sections spanning several hundred meters within unconventional reservoirs. To address these shortcomings, we introduce a novel spectrum enhancement framework, the R/S-STFT method. This approach innovatively integrates the foundational principles of R/S analysis with the short-time Fourier transform (STFT), a time-frequency analysis (TFA) technique. The workflow involves generating R/S curves from preprocessed and reconstructed conventional logs. Instead of applying the FD method, the STFT is utilized to transform these curves into the time-frequency domain. This transformation allows for the precise extraction of the Fourier spectrum and the strategic filtering of high-frequency noise, which often obscures the concave segments on the R/S curve that are indicative of fractures. The filtered signal is subsequently reconstructed into a smoothed R/S curve using the inverse STFT, facilitating robust and precise detection of concave positions. A composite fracture probability index is then derived by synergistically integrating extracted spectral characteristics with waveform morphology within the identified depressed sections. The proposed R/S-STFT method was validated using a comprehensive data set from Horizontal Well P1 in a tight sandstone reservoir of the Ordos Basin, China, incorporating entire wellbore image logs and corroborated by independent drilling engineering data. The results demonstrate superior performance over the traditional R/S-FD method, which achieved only 59.52% recognition rate with 67.57% accuracy, while the R/S-STFT method attained a total fracture recognition rate of 70.21% and a notably high accuracy of 89.19% within the identified fracture zones. Furthermore, the R/S-STFT method significantly reduced false positives in nonfracture areas, and the resulting fracture distribution was consistent with independent geological knowledge. This study establishes a feasible and effective application of TFA within a fractal theory framework for subsurface fracture identification, offering geophysicists a reliable tool for enhanced fracture modeling and informed fracturing operations.
Liu et al. (Fri,) studied this question.