Cross-Well Lithology Identification Based on Wavelet Transform and Adversarial Learning

For geological analysis tasks such as reservoir characterization and petroleum exploration, lithology identification is a crucial foundational task. The logging at this stage generally build model, assuming that the data share an independent identical distribution. This assumption, however, does not hold among various wells due to variations in depositional conditions, apparatus, etc. In addition, current model fully integrate knowledge, meaning geologically reliable easy interpret. Therefore, we propose cross-domain method incorporates information domain adaptation. consists of designing named UAFN structure better extract semantic (depth) features curves, introducing via wavelet transform improve model’s interpretability, using dynamic adversarial adaptation solve data-drift issue cross-wells. experimental results show that, by combining coefficients with information, more lithological can be extracted curve. Moreover, performance further improved transform. addition average 6.25%, indicating value stratigraphic contained for prediction.