Shear wave velocity prediction based on gradient boosting model enhanced with correlation features

Well log data represent a unique form of tabular data commonly used in geophysical and geological analyses. Mining the underlying relationships among logging parameters is crucial for downstream tasks, such as shear wave velocity (S-wave velocity) prediction, which is a key focus in oil and gas exploration and development. Gradient boosting models are widely recommended for regression tasks involving tabular data because of their strong predictive capabilities. This paper proposes an enhanced gradient boosting framework that incorporates correlation features for more accurate S-wave velocity prediction, thus improving the accuracy and generalizability. Using the volcanic reservoir in the Xujiaweizi Depression of the Songliao Basin in Daqing as a case study, we develop a correlation feature extractor tailored for well log data to leverage the full potential of feature engineering. The extracted correlation features are then integrated with raw data to further enhance the prediction performance of gradient boosting models. Experimental results demonstrate that models augmented with correlation features outperform unenhanced gradient boosting models and deep learning models.