Abstract
Classifying natural gas hydrate reservoirs effectively and carrying out reservoir classification modelling is crucial, but to date, research on building artificial intelligence-assisted logging curve reservoir classification models is not abundant. As exploration and development have progressed, an increasing number of fine-grained reservoirs are being discovered, and their strong heterogeneity makes correct reservoir classification even more important. Two wells used for detecting hydrates in the Qiongdongnan (QDN) Basin are used to explore the relationship between logging response parameters and reservoir quality, as well as the method of building a logging-based reservoir classification model. Through K-means clustering and Adaboost methods, the K-means method is considered to be able to correspond to the hydrate enrichment degree, while the random forest method can establish an effective reservoir classification model (the recognition accuracy is 95%). In the different categories of reservoirs, the physical properties of the reservoirs are obviously poor, and the corresponding hydrate saturation is also low, which indicates that heterogeneity has indeed affected the enrichment of hydrates in fine-grained reservoirs. This reservoir classification research method can effectively recognize reservoirs.