Abstract: In recent years, some exploratory wells in the western South China Sea oilfields have successively shown gas logging anomalies characterized by high heavy hydrocarbon fractions and low light hydrocarbon fractions. To address the issues of unclear understanding of the characteristics, unclear formation mechanisms, and difficulties in fluid property identification for such high heavy hydrocarbon fraction gas logging anomalies, statistical analysis was conducted on the gas logging fraction anomaly data in the study area to identify the causes of these anomalies, and research was also carried out on the identification of fluid properties using machine learning classification and prediction models. The analysis suggests that the main reason for these gas logging anomalies is the stimulation or destruction of oil reservoirs after their early formation, leading to secondary migration of oil and gas, during which light hydrocarbon fractions are lost in larger amounts, while heavy hydrocarbon fractions are lost less. The heavy hydrocarbon fraction gas logging anomalies are influenced by a combination of factors, including the degree of geological structural changes, the sealing properties of the cap rock and the properties of the crude oil. The random forest model demonstrates good training and predictive classification performance on the high heavy hydrocarbon fraction gas logging data set in the study area accurately and efficiently identifying fluid properties. This study provides a new approach from the perspective of gas logging technology for identifying reservoir stimulations and has significant guiding implications for oil and gas exploration and development.

Key words: gas logging, hydrocarbon fraction, gas logging fraction, secondary reservoir, machine learning