关键词: 测井曲线重构, 随机森林, MDA算法, 机器学习, 特征重要性

Abstract:

Logging data plays an important role in oil and gas exploration and development, and is widely used by geologists to address various geological issues. However, due to instrument failures, operational errors, or complex geological conditions, logging data may encounter problems such as missing parameters or incomplete records. Re-logging is costly and constrained by scheduling factors, making research on logging curve reconstruction based on machine learning methods highly significant. This paper proposes an improved random forest (RF) model based on the MDA (mean decrease accuracy) feature importance algorithm (MDA-RF model) for logging curve reconstruction. The MDA algorithm is used to assess the sensitivity of input features to prediction accuracy, removing features with low relevance to the target parameters, thereby enhancing the predictive capability of the RF model. Experimental validation was conducted using actual logging data from the Huizhou region of Guangdong Province. The results indicate that the MDA-RF model improved the coefficient of determination by 1.6% and 5.0% when predicting individual wells and completing missing well intervals, respectively.

Key words:  , logging curve reconstruction, random forest, MDA algorithm, machine learning, feature importance